TW484104B - Apparatus and method for the inspection of objects - Google Patents

Abstract

A method for inspecting objects, the method including creating a reference image for a representative object, the reference image including an at least partially vectorized first representation of boundaries within the image, acquiring an image of an object under inspection including a second representation of boundaries within the image, and comparing the second representation of boundaries to the at least partially vectorized first representation of boundaries, thereby to identify defects.

Description

484104 A7 V. Description of the invention The invention relates to an apparatus and method for image processing. BACKGROUND OF THE INVENTION The apparatus and method for analyzing images, especially the apparatus and method for processing and analysis facilitate the inspection of patterned objects It is well-known in related arts. The following references show that image processing methods can be beneficial to the invention: ♦ C. Gonzalez and P. Wintz, AddiSon Wesley Publishing, Reading, Massachusetts, 1987, "Digiul Image prOcessing ,, and The Image Processing Handbook, by John C. Russ, published by CRC Press in 1994. The following references describe edge detection methods that may be helpful in understanding the invention: D. Marr and Published by E. Hildreth in the Proceedings of the British Academy of Sciences, "The 〇Q of Edge Detection"; and "A new chromatic edge-detector used f〇" published by the International Pattern Recognition Conference held by M. Chapron on April 11. r ima O ^ segmentation ”0 The following references describe color image segmentation methods that may be helpful in understanding the invention:

The 7th International Advanced Machine held by Philippe Pujas and Marie-Jose Aldon in San Filiu de Guixols, Spain on September 22, 1995. (Please read the precautions on the back before this page) Order: Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Online Economics -4- 484104 A7 V. Description of Invention (2) "Robust Colour Image Segmentation" published by the conference of robots ,,;as well as

Lelia Shararenko, Maria Petrou and Josef Kittler, International Society of Electrical Engineers, Automatic Watershed

Segmentation of Randomly Textured Colour Images ... U.S. Patent Case No. 4,758,888, which belongs to Lapidot, describes a method and apparatus for inspecting work in a traveling production line, including a method and apparatus for inspecting defects online without interrupting progress along the production line or work. U.S. Patent No. 5,586,058 and No. 5,6 1 9, 4 2 9 belonging to Aloni et al. Describe a method and apparatus for inspecting a view-type object and detecting defects, including a binary level representation of the inspected object , View the gray level representation of the object 'and better re-view the gray scale representation of the object to filter false alarms and classify defects. U.S. Patent Case No. 5,774,572, which belongs to Caspi, describes an automatic visual inspection system. 'The system operates as a 2-dimensional digital grayscale scale image of a convolution object with a filter function related to the second derivative of the Gaussian function, forming an image with Two-dimensional convolution image of sign 値. The position of the edge of the object is achieved by looking for a zero crossing between adjacent opposite signs 値. U.S. Patent Case No. 5,774,573, belonging to Caspi et al., Describes a visual viewing system 'this system uses a convolution with a 2-dimensional digital grayscale scale image of an object that is related to the second derivative of the Gaussian function, and A two-dimensional convolution image with a sign 値 is formed. The convolution of Caspi et al. Can be performed using two Gaussian differences, one of which is positive and the other negative. PCT application No. IL98 / 00393 illustrates the use of color The printing-5 private paper size applies the Chinese National Standard (CNS) A4 specification (210 X 997 public multi-factory MI installation --- (Please read the precautions on the back before this page) Order · · • Line-Intellectual Property of the Ministry of Economic Affairs Printed by the Consumer Cooperative of the Bureau A7 B7 V. Description of the invention (3 View of the circuit board, including the use of color to identify specific types of conditions such as the oxidation of electrical conductors using color. PCT application No. IL98 / 00477 explained analytical formula The method of representing images, so that type operations such as dilation, erosion, and scaling can be performed on non-binary pixels, which is more efficient and effective than using previous methods. The listed patent application No. 125929 describes an article inspection system and method, including an improved image acquisition system. Methods that can facilitate image analysis are described in the following published literature: NJ 08855 Piscataway, Rutgers University, New Jersey, USA Computer Engineering Department, written by guest Dorin Comaniciu and Peter Meer in spring 1998, Distribution Free Decomposition of Multivariate Date ". The disclosure of all references mentioned above and the entire specification, as well as the disclosure of the documents cited thereby, are hereby mentioned in the manner mentioned. Included herein. _ SUMMARY OF THE INVENTION The present invention seeks to provide an improved device and method for image processing. The present invention facilitates the combination of the Israeli patent application No. IL 13 1092 filed by the applicant on July 25, 1999 No., named "Optical Inspection System", the disclosure of which is incorporated herein by reference. The inputs of the image processing system shown and described here typically include at least the following inputs One of them: a. Selectively, the area of interest or the speed of the area Photos, including some or more of the following areas: areas pre-defined by the user, around the suspected defects-6-[Paper size stab towel family standard (CNS) A4 specification (21G X 297 public love) '' 1 ^ ^ ------- install --- (please read the precautions on the back before @this page) • trDV · _ Line · Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 1U4 A7 B7 V. Description of the invention ( 4 The fields printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs and the fields around the characteristics of predetermined types; b • binary CEL data (sometimes referred to as ncel data π), which is equivalent to identifying between different areas of the figure Data of peripheral elements of the boundary; c • Color CEL data, called data identifying peripheral elements equivalent to the boundary between different areas of the figure and data identifying the type of area on each side of the CEL; d. Triggering of type feature viewing; e • Color defect inspection trigger; f · Binary defect inspection trigger (sometimes referred to as "hard defect"), which is a nick, protrusion, and CEL on the skeleton, where CEL is on the type skeleton; and g · User identification View trigger areas of interest. Preferably, these inputs are provided by the image processing unit in essence, as described in the applicant's Israeli patent application No. IL 131〇92. Therefore, there is a method for inspecting an object as provided by a preferred embodiment of the present invention. The method includes creating a reference image of a representative object, the reference image including a first representation of at least a part of a vectorized inner boundary of the image, obtained in The image of the object under review includes a second representation of the boundary within the image and a comparison of the second representation of the boundary with the first representation of at least a portion of the vectorization of the boundary due to 'this identification defect. In addition, as in this specific embodiment of the moon car, the comparison step utilizes a user-selectable variable setting to be limited to the acceptable distance of the corresponding portion of the boundary between the ^^^^^ and the second representation. . In addition, it has the same paper size as the preferred embodiment of the present invention, which is used for the image paper size Gu + Guan Jiaxian ------- I ------- install --- (Please read the back (Notes on this page) Order.-· Line · V. Description of the Invention (5) The processing system includes a boundary identifier, which operates to generate a representation of the known element boundary in the image, and a hardware candidate defect identifier, which operates to The hardware candidate defect candidate is recognized in the image, the software candidate defect viewer receives the output from the hardware candidate defect identifier, and the software uses a boundary representation to identify at least one false alarm within the output. Preferably, the at least one false alarm acts by comparing the boundary with a reference. In addition, as in the preferred embodiment of the present invention, the boundary identifier includes a hardware boundary identifier, which operates to generate a table of known element boundaries of the image in hardware. 7 In addition, as in the preferred embodiment of the present invention, the system It also includes a software candidate defect identifier, which operates to identify the remaining candidate defects in the image with software. Also, as in the preferred embodiment of the present invention, the software candidate defect viewer also receives a second output by the software candidate defect recognizer and expresses it as a software use boundary < to identify at least one false alarm in the second output. . In addition, as in the preferred embodiment of the present invention, the hardware candidate defect recognition uses a representation of a boundary to identify at least some candidate defects. In addition, as in the preferred embodiment of the present invention, the software candidate defect recognition uses a representation of a boundary to identify at least some candidate defects. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. In addition, as provided by the preferred embodiment of the present invention, the system for image processing includes a learning sub-system, which operates to define the first and second fields of objects under inspection. The first field each contains at least one known critical object element and the second field does not include the known critical object element, and a defect detector operates to use the first based on prior knowledge of known critical object elements. Procedures to review the first field, and use a second procedure different from the first process -8-484104 A7 V. Invention Description (6 Intellectual Property Bureau of the Ministry of Economic Affairs Employee Consumer Cooperatives to print a second procedure to review the second field. In addition, such as the present invention In a preferred embodiment, the second package includes hardware inspection in the second field, which operates to identify candidates in the second field and a subsequent software view that includes the second field. Only if at least one is found in the second field When candidate defects are detected, the operation is to analyze the at least one candidate defect found in the second field in order to identify false alarms. Also, as provided by the preferred embodiment of the present invention, the method for defying the object includes identifying the position of the at least one candidate defect in the first stage of the inspection, and for each candidate defect, The position < candidate field decision is used to review at least partly the size and shape of the peripheral region based on the output of the recognition step, and in the second stage of the review, review each candidate field to confirm the defect. In addition, as in this The preferred embodiment of the invention is made. In addition, the preferred embodiment of the invention is made. In addition, the preferred embodiment of the invention is the size of a candidate field in a known position. In addition, the preferred embodiment of the invention is implemented as a preferred embodiment. For example, the candidate field or the criteria for the candidate defect are implemented differently. Also, as the preferred embodiment of the present invention, the second stage includes execution based on the characteristics of the at least candidate defect identified in the identifying step; The same review. In addition, if the present invention is a preferred embodiment, the second stage includes completing the decision step in hardware and the second step in software according to the first stage. Step Inclusion Decision This second stage includes basis (Please read the precautions on the back before HP: this page)-Packing price. Thread. -9- A7

Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. The function of the resident part of the candidate defect performs different inspections. In addition, 'if the present invention preferably has an embodiment, the second stage includes the selection of the function of the resident part of the defect and the degree of urgency to perform different inspections. Two == Provided by the preferred embodiment of the present invention, it has the use of A modular image processing system with artifact analysis functions is used in combination with disk preparation; the system includes an image processing engine, which scans and cries: ^ at least-image data stream to be analyzed, and contains-engine construction consulting As a receiver, one of the user-customizable image analysis functions is to be executed on the image data by the image processing engine. Sequence = The image processing engine operates as a sequence of feeds fed to the engine builder. I define and analyze at least one. The channel image resource section includes the execution of different image and analysis functions, and it is sufficient to reach the engine builder based on the sequence definition, not only the relevant parameters are different from each other. It is also 'as provided by the preferred embodiment of the present invention. The optical inspector method ΓΓ is included in the first inspection step and defines the number of regions of interest used for image processing. The region of interest includes The user-defined at least one region of interest and the at least one region of interest that are automatically defined by viewing the object in a scientific manner, and provided to the image information associated with the area surrounding each region of interest; The information is used to determine the lack of the object automatically processing the region of interest. In addition, if the image of the preferred embodiment of the present invention is smaller than the image of the object. In addition, if the preferred embodiment of the present invention surrounds the region of interest, the object is viewed optically. and

This paper rule size_ tm (CNS) A4 (2I0 x 297mm) Μ 7 Μ -------- Order ----- {Please read the precautions on the back before you page}

ϋ I n I ^ ---- 484104 A7 B7 、 Explanation of the invention (8 The area of interest that is automatically defined includes candidate defects for the formation of graphics on objects. --- "17 ------- install- (Please read the notes on the back first and then this page): Outer 'As the preferred embodiment of the present invention, the area of interest is automatically defined by optically inspecting the object' contains the predetermined pattern features formed by the graphics on the object Also, as in the preferred embodiment of the present invention, the providing step includes providing analysis information about the boundary of the region of interest to the image processor. Preferably, the providing step additionally includes providing a color image of the region of interest to the image processor. Image processor. In addition, according to a preferred embodiment of the present invention, the automatic processing step includes applying at least one image processing method to an area of interest defined by a user, the vhai method, and application to optically inspecting an object. The image processing method of the region of interest that is automatically defined is different. • Line-Also as the preferred embodiment of the present invention, the providing step includes identifying the type of defect in the region of interest, and the automatic The processing step includes applying an image processing method suitable for the type of defect in the region of interest. In addition, according to a preferred embodiment of the present invention, the providing step includes identifying a type of complaint feature type in the region of interest, and the automatic processing The steps include the method of printing and applying an image processing method suitable for the type feature type in the region of interest by the Intellectual Property Bureau employee consumer cooperative of the Ministry of Economic Affairs. Also, as in the preferred embodiment of the present invention, at least a user-defined at least A region of interest, at least one region of interest that is automatically defined by optically viewing the object before the optical inspection of the object in the software definition step is performed in the hardware inspection step, and the images of the region of interest are Automatic-11 A7 V. Description of the Invention (g) Processing is performed in the software image processing step. In addition, if the present invention is better and better, the present invention provides a method for inspecting a ball grid array substrate. , Table 4, and u 40,000 to include the reference to generate the spherical grid array substrate ~ the pattern of at least-features, and to store the pattern To the memory, obtain the image of the ball grid array substrate, and view the predetermined area of the image of the ball grid array substrate with the order & I wonder whether the characteristics of the pre-footed area are suitable for this mode. In addition, as this In a preferred embodiment of the invention, the feature includes a circle. In addition, as in a preferred embodiment of the invention, the mode of the circle includes the center point of the pre-position and the radius within a predetermined tolerance. In the preferred embodiment, the parameters of this mode can be adjusted at least partially in offline mode before viewing. Also, 'as in the preferred embodiment of the present invention, this feature includes _ pads. Brief description of the drawing The present invention will be explained in detail by the following It is understood and recognized with reference to the accompanying drawings, in which: FIG. 1A is a simplified block diagram of an image processing system, which is constructed and operated as a preferred embodiment of the present invention; FIG. 1B is a simplified block diagram of an image processing system. FIG. 2 shows the construction and operation of another preferred embodiment of the invention; FIG. 2 is a simplified functional block diagram of the device of FIG. "Design rules" mode; Figure 3 is a simplified functional block diagram of the unit 2 1 0 of Figure 2; Figure 4 is a simplified functional block diagram of the device of Figure 1, which is used by the user at -12- $ paper scale Detailed Zhongguanjia Standard (CNS) A4 specification (21 (3x 297 public ίΠ " --m ^ equipment --- C, please read the precautions on the back before this page).-Line _ Employee Consumer Cooperatives, Intellectual Property Bureau, Ministry of Economic Affairs Print 484104 A7 V. Description of the invention (10) The selection scheme is the "learning" mode; Figure 5 is a simplified functional block diagram of a single unit of the "mixed learning" unit in Figure 4; Figure 6 is a block diagram of Figure 5 A simplified functional block diagram of a single block that is better made of the three blocks; Figures 7A-7B, taken together to form a better test function table that explains the types of windows provided in the study plan course; Figure 8 is a multi-tasking task from Figure 5 The best function table for the typical implementation of the cut-off manager in the learning plan course; Figure 9 is the implementation diagram of the knowledge plan, which can be, iearn ref c0nfig ,; Figure 10 is the execution plan of the scanning plan, from This is "learned.ref.eg.confign; Figure 11 The execution diagram of the scanning scheme is set to inspect.mixed.config "; Figure 12 is the execution diagram of the scanning scheme, which is set to M inspect.mixed.egmain.con fig'1; Figure 13 is the execution diagram of the scanning scheme Therefore, it is set to 丨 丨 inspect, mixed .egside.config 丨 丨; Figure 14 is the execution diagram of the scanning scheme, which is set to " inspect.mixed.eg.config " i Figure 15A-15C, taken together to form Explains the better test function table for each window type provided in the view plan course. Figure 16 is the best function list typically performed by the multi-cut manager of the work shown in Figure 12 in the view plan course; -13 This paper standard applies China National Standard (CNS) A4 Specification (210 x 297 mm). It-Packing --- (Please read the precautions on the back before this page) · Line. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 484104 A7

Figure 17 is a better data structure diagram of the report component. It can be analyzed by the Israeli patent Shen Jing'an which is still under trial, and the hardware of the China A% Yimu Bluff as the younger brother 13 1092. Figure 1 8 is a better data structure of the header of FIG. 17; FIG. 19 is a simplified execution diagram of the image processing operating system sip of FIG. I; ☆ FIG. 20 is another example of the simplified execution diagram, which is different from FIG. 19 and is adapted Horses are suitable for execution by a computer consortium that includes at least two simultaneous parallel processors; Figure 21A contains a conceptual explanation of a CEL pixel; Figures 21B-21D are relationships between features and analytical expressions generated by muscle and vectorized line elements Explanation of concepts; Figures 22A-22H are explanations that are preferably used to define edge codes that are included in a single record's edge type; Figures 23A-23H are explanations that depict edge codes 为 0 to 9 and grid orientation respectively; ° Figures ui is the interpretation of CEL defined by the following parameters: edgec〇de = 〇, first = 14, last = 〇; Figure 24 is an image interpretation of the envelope data structure surrounding CEL; Figure 25 is an envelope that encloses two consecutive line segments respectively Image interpretation; Figure 26 is in Simplified image interpretation of aggregates of equal frames collected on the viewing panel of a preferred embodiment of the present invention; Figure 27 is an image interpretation of an example of data that may exist in a window; Figures 28A-28B are parallel and non-parallel, respectively The best form of image interpretation for line width measurement; ------- ^ -------- install --- gun (please read the precautions on the back before this page) —line- Printed by the Employees' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs -14-484104 A7 B7 V. Description of the invention (12) Figure 29 is an image explanation of the foot-to-foot conversion of the bismuth pressure system between the online and reference seats; For the image interpretation of the vectorized object with the envelope, and the reference to the edge shape, each G frame is related to the periphery of the edge and the allowed margin difference limit. Figure 30B is a polygon reference superimposed on the figure. The image interpretation of Cm online has Amami registration between online and Zhu Kaozi, Daddy Beibei, and it shows defects;

FIG. 30C is an image explanation of the line cel superimposed on the polygon of FIG. 30A. Μ ^^^^^ Refer to the image explanation of the line cel with 'between the line and the reference material>, 1, 3, and 10 only. The defects that are not displayed are false alarms and other defects. Figure 31 is a flowchart of a better method suitable for micro-temporary storage of reference images and online images in the viewing window. Line) Simplified image interpretation of CEL aggregation in the referenced micro-registration; Figure 32B is an image interpretation of the CEL aggregated by its reference registration after micro-registration according to the preferred method of Fig. 31; line diagram 33 is a process Figures show the functions of an image processing system constructed and operated as another specific embodiment of the present invention; and Figures 34-42 are screen systems generated by a general user interface for building and constructed by a specific embodiment of the present invention and The operating image processing system defines the viewing parameters of the image. Detailed description of the drawings In the following specifications, although for the purpose of explaining the present invention, references are extended to the ball grid array substrate (BGA), it is easy to recognize that the present invention can be applied -15- A7 B7 V. Description of the invention ( 13) For any suitable graphical object. Therefore, the BGA of this project is not limited to the ball grid descending substrate alone, but is considered as a reference to include any suitable form of patterned objects = p = road board. Substrates, layered printed circuit boards, guidewires, flat R-bonded chip package substrates, tape-shaped auto-welding substrates, and any suitable: = ㈣ and, for example, a rigid metal substrate can be used. The figures shown in the figure are not otherwise indicated. The squares represent functional units and the circular squares represent data processed by the functional units. : Referring to FIG. 1A, this figure is a simplified block diagram of an image processing system, such as the construction and operation of one of the preferred embodiments. In this specific embodiment, the scanner 2 preferably represents the object 4 by optically scanning the object * and generating an RGB (red, green, blue) two-dimensional image, and outputs RGB image information to the image processor 5. It is preferable to provide a domain identifier of interest 6 and the image processor $, the space of interest identifier 6 and the line width and frame processor 3 output additional viewing triggers and other SIp units 7 located downstream, The execution of this unit is not provided by Shadow = Provision 5 (extra viewing function, preferably the viewing function is adapted to the viewing trigger input, so that various types of viewing triggers "are the areas that are relevant to the trigger to be most suitable for viewing the The inspection routine of the trigger area is used for inspection. Scanner 2 is preferably an image illumination and acquisition system, as explained by the applicant in parallel. J t It t -Illumination for Inspecting Surfaces of hUc 丨 es ". Although Scanner 2 is more Best operation is to get rgb light (please read the note on the back first-installation _ matters and then this page) M Ministry of Economic Affairs Bureau Intellectual Property Bureau employee clothing cooperatives printed clothing ^ --------- ^ ---- -------------------- -16- V. Description of the invention (14) It is easy to recognize that the scanner can operate, such as monochrome grayscale images, H The image related to the electric field of the form image, Xi Emperor: Bei Ruyuan wrapped around the printed circuit board The special electrical properties related to the surname and private property of the printed circuit board are related to the thunder path on the printed circuit board. It is recognized that the present invention can be related to the image processing of optical or non-optical images. Embodiment 'The image processor 5 is preferably operated to receive ~ Like', "analyze image information resources to detect and detect suspected defects and downstream downstream '," Like "and its video recording according to MB 2 as early as 70 The predetermined features are only available for triggering. The image processor 5 is compared with the following modules: ^ The remaining sets of processing orders S, including the ⑴ binary CEL recognizer 8 module. The edge between the two 4 boundary areas = the arc horse is not introduced The boundary of Yu Wujia (peripheral components, where the high gradient bit t bar image is between the darker and brighter areas. Two: the system is hereinafter referred to as "CEL, or ,, binary cel.,. Binary-recognition The scanner 8 receives and processes the image data from the scanner 2 and outputs a binary CEL report that exists in the image of the four-shirt image of the object. (Π) The color CEL identifier 100 〇 r / color CEL indicates that it is in the boundary area The periphery of the edges between different color groups in the object shape. The color groups are the same The color area, the color c EL material, is customized by Xu M and is better related to the material. ”The color group selected between the right color group, identifies which color group exists on each side of the edge: Recognition = ㈣Jia is to process the image data received by the scanner 2 to identify different color groups in the image data, related to 484104 A7 B7 known in the object (5) Description of the color group of the material, and Wheel + Output the color CEl report of the image existing in Object 4: to define different materials and the boundary between the regions. Pre = two terrorists were identified by the 11. The type feature is a premonition that can be identified in a figure. The type special Wei ## 我 我 别 4 preferably works as received by the scanner 2-at least-input, the binary CEL recognizer 8 receives the primary CEL information from Yers ~ Ling ^ Λ and by The shirt color CEL recognizer 10 receives the color feature report i. The special break identifier U processes these inputs and the output type = report 'triggers a further review in the downstream review unit. It is preferred that fewer form features be related to the material in the object 4. In the text of this article, "type features include, for example, conductor stubs, openings, shorts, joints, and other similar features. Alignment ^^ Detector 13. The defect is a predetermined type that can be identified in the graphic: day. The defect identifier 13 operates as image data generated by the scanner 2, the binary CEL identifier 8 and the color identifier 1G. At the very least, the defect identifier 13 processes these inputs and outputs suspected defects to trigger additional inspections in the tour inspection unit. . Preferably, the defect is identified by the defect =: U reported as a reference to a set of predetermined design rules = the input data is not compared with the reference. In the BGA inspection, the K defect includes, for example, the definition of unevenness on the periphery of the conductor. The predetermined parameters of flat, oxidation, low-contrast color surface defects, such as chemical residues of I, and tritium-contrast surface defects, such as scratches. (V) Snap action generator 14. When present, the snap action generator μ is preferred The ground operation is triggered in response to the reception by the pattern feature identifier U and the defect recognition cry 13 and the defect to generate the selected field in this image: snapshot L_________ -18- This paper is applicable in the standard (CNS) A4 size coffee 297 public ^ 484104 A7 B7 V. Description of the invention (16) image. The snapshot image is a part of the image received by the scanner 2 and is preferably generated around the type feature identifier u And the triggers received by the defect recognizer ，, the window of interest. In addition, the fast-moving generator μ can generate the space-of-interest area < fast-moving image according to the trigger received by the space-of-interest area recognizer 6 < Detailed description. It is preferred that the shape and size of the fast-moving image output by the fast-moving generator 14 is structured according to a specific type of region of interest in response to triggering. The fast-moving generator 14 preferably operates as The output image data is easy to be converted into downstream processed dynamic images. The hardware suitable for the image processor 6 as the image processor 6 is described in the applicant's pending patent application No. ^ 13i092 It is easy to recognize that the above description is composed of the image processing functions existing in the image processor 6. However, the functions of the image processor can actually be classified into modules, such as two Processing, color processing, and type processing, and issue functional reports classified as described above. The printed reports of the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs can be classified into various reporting channels based on the hardware architecture. Therefore, in the specific embodiment of the present invention "The image processor 5 generates reports provided to SIP 7 in two channels", where each report channel includes a number of reports. It is preferably "No .: channel, channel, designated as " fast-moving channel " Device 13 generates (a report of defect data related to color surface defects, and a reporter including a fast-moving image of a region of interest generated by the fast-moving generator 14. The second channel, M is a "CEL " channel, preferably Including the binary CEL report generated by the binary recognizer 8 and the color CEL generated by the color CEL recognizer 1G, called the defect report to report, and generated by the pattern feature 11 (type features and also by the defect recognizer 丨The two produced by 3-丨 σ-This paper size is applicable to China National Standard (CNS) A4 specification stone 10 × 297 male 19

Defects such as scores and protrusions. i Zhuangyi --- (Please read the notes on the back first, then this page) The preferred embodiment of this month, the 6-edge identifier of the space area of interest is for the stable. I repeat the sense in a viewing object. Areas of interest. Existence: The space area of interest in Object 4, a particular type of feature is expected to emerge, and the pre-defined routine trimming is a typical execution type feature. I, preferably, the space-of-interest generator 6 learns 1-style operation offline before defect inspection, which is explained in more detail below. In the better and cleverer :: medium: the space-of-interest area identifier 6 includes two modules: make sense < field-of-interest module 8 and computer-generated field-of-interest module preferably, during learning In the mode, the following is explained in more detail through a general user interface with reference to FIGS. 34_42, which is a part of the user-defined interest area module 8. A general area of user-defined view interest is located in the object batch to be viewed. Of objects 4. Preferably, the user also identifies a type of structure that is expected to be located in the area of interest. Therefore, in this article reviewed by BQA, the user can define general areas including pads, targets, wafer areas, or spheres. In the Intellectual Property Bureau Employee Consumer Cooperative of the Ministry of Economic Affairs, private computer-generated modules of areas of interest 9 are printed. A computer processor can be included in the SIP 7 image analyzer 2 4. • 'Learning η is identified by the user as the location of the structure of interest within the spatial domain defined by the user in the user-defined interest domain module 8. The computer-generated domain of interest module 9 preferably produces a specification that closely surrounds the localized space domain of each structure within the domain defined by the user. Therefore, in the BGA inspection -20- ^ paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 public love) ---- 484104 A7 B7 V. Description of the invention (18) Employees ’Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs In the printed text, if the user defines the general area including the welding pad, the computer-generated area of interest module is defined by the user to define the structure and position of each pad, and whenever the welding pad is expected to be positioned ^ Interesting modules generated by Zhongyu Computer define localized areas of interest. All the structures surrounding the user's foothold are stored in the memory, which are related to the SIP 7 (not shown), and then accessed in the view. X 1 is a preferred embodiment of the present invention, and SI ”is provided by the downstream processor 5 and operates to receive a video by the image processor 5 and execute a routine not provided by the image processor 5. For example, Up • Part two, to provide a comparison mode view, in which the data generated by the object 4 being viewed is compared with the reference data of the object. In addition, the SIp7 part operates to be related to the stored information generated by the space-of-interest domain identifier 6 based on the two prior The definition of the rules is based on the dynamic adjustment of the inspection routines for the types of features detected by the image processor 5 during the imaging process and the types of defects detected by the image processor 5 during the imaging process. Preferably, SIP 7 is based on software. The benchmark image processor executes a custom software image processing routine on a workstation such as a microcomputer workstation running in a linuxtm environment, including the following modules: ⑴ line width and processor module 3, (ii) work shrink Integrator and interested window generator module 12 and (iii) analyzer module 15. In the preferred embodiment of the present invention, the line width and registration processor module are received by a binary CEL identifier 8 CE] L and the type feature data received by the type feature detector 11 and processing of input information in the following modules: (1) The 5 book generator module 18 operates to execute the installation in the conversion calculator 20 Special guest note: The paper size is applicable (cns) a4 specifications ⑽χ 297 21 484104 A7 B7 V. Description of the invention (19)? The conversion of the main book is based on the patient's book in the examination of the heart / image ~ reference image and in the The excess / leakage count in the nose device 2 2 determines the presence and absence of the 5th and 3rd days to determine whether all the expected characteristics are present in the inspection man # 1 & contempt object 4; and ( ii) The line width calculator 16 is operated by the conversion calculator 2 to receive registration conversion data to calculate the line width of the conductor and verify that the line width at each particular location is within the allowable range of that location, and verify that it is between the conductors The space between them is greater than or equal to the predetermined total 値. ^ 軚, the construction and operation of 1 booklet generator 18 and its related conversion calculator 20 and surplus / missing calculator 22 are explained in more detail below. Invention-the preferred embodiment, the work of SIp 7 The reducer / window of interest generator unit 12 preferably receives inputs from the following modules: binary identifier 8, color CEL identifier 1 (), pattern feature identifier 11, defect identifier 13, conversion calculation Device 20, excess omission calculator 22, and line width calculator 16. In addition, the work reducer 12 receives the space area of interest defined by the interested jade field recognizer 6 in the learning mode and then stores it in the memory. . The Consumer Property Cooperative Print Job Reducer 12 of the Intellectual Property Bureau of the Ministry of Economics operates to generate interesting windows that trigger around various views. The view trigger is preferably a predetermined type of feature received by the pattern feature recognizer n, a predetermined type of defect received by the defect recognizer 13 and a predetermined space of interest area identified by the space of interest area identifier 6 and a line width and The predetermined line width and excess / missing defects received by the registration processor 3 are stored in the memory. Optionally, each view trigger is provided as a snap motion image generated by a snap motion generator 14 triggered around S. Preferably, the shape and size of the fast-moving image are dynamically responded to the trigger type according to a predetermined rule. 22 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm). Α7 Intellectual Property Bureau staff, Ministry of Economic Affairs Printed by Consumer Cooperatives V. Invention Description (20) Type and / or Location. For each view trigger, the work shrinks.] 9,! M, and the title is prepared.12 It is better to generate the identification including the dynamic selection of the appropriate view task in response to the received special issue to perform the special view task, and to perform Appropriate image information is required for visually subscribed jobs. For example, in the text of the BGA review, the package of the signing field of the space of interest field preferably includes the identification with binary CE; D color CEL and 2 area type features date $ m Μ welding field review. The package identified as zone 2 with colored surface defects preferably includes the identification of the inspection of the colored surface defects and the area of interest (quick-motion image, but without CEL information. Areas identified as having nicks / crown defects) The package preferably includes identification of the appropriate notch / protrusion inspection method, binary cavities of the affected area, and optionally a snap-action image. The job reducer 12 typically transmits dynamic processing components, including image and method information ' It is indicated by the big arrow to the analyzer unit 15. The dynamic information is preferably = the analyzer 15 that dynamically transmits a specific defined area of the image based only on the defects, features and information of interest received by the job reducer 12. This, preprocessing " Obviously, the amount of image data analyzed by the analyzer unit 15 is low, and therefore the calculation load is reduced, thereby greatly reducing the cost of the inspection room related to the image analysis of the object. A specific embodiment of the present invention, the analysis The analyzer unit 包括 preferably includes a unit · an image analyzer module 24 and a post-processor module 26. ~ The image analyzer module 24 preferably operates to receive the Dynamic processing components and analysis by using the method identified by the job reducer 12 (CNS) A4 specifications ⑽ -23- X 297) (Please read the precautions on the back before 0 pages) '丨 装-\ ^ τ.. Line. V. Description of the invention (21) ===, identified by the work reducer 12 and imaged: A. Viewing work based on the comparison of Wanfama, among which,. =: See: The area in the image is analyzed as well as the reference image. It is better to compare the reference images based on the CEL spot in the image of the object 4 being inspected and the area related to the trigger area of the suspected defect. The post-processor 26 preferably operates as a defect-producing device, which is described below. It is also preferably operated as a reference to analyze the snapshot image of the area scratched by the image analyzer with a foot scratch to include A predetermined type of defect, in which the suspected defect has a high probability of being a false defect. SIP 7 is actually used as described here, and it is preferably provided based on the interesting window provided by the sense of interest, fun and self-generation Defect output report 30 and object features. &Quot; Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs Now refer to FIG. 1B, which is a simplified functional block diagram view of a BGA (ball grid array) substrate inspection system, as shown in the present invention. The construction and operation of another preferred embodiment. The BGA viewing system of FIG. 1B is similar to the preferred embodiment described with reference to FIG. 丨 A, however, the BGA viewing system shown in FIG. 1B is significantly different and the display is sensed by 3 The input from the camera or the camera is part of the scanner 34 (equivalent to scanner 2 in FIG. 1A), and does not include the snap action generator 14. The view is that the view system of the Bga view is The reference is for explanatory purposes' and any suitable object can be viewed. 32 It is believed that the viewing system shown in Figure 1B includes a user workstation. The scanner 34 typically produces RGB images or is intended to be viewed. The other numbers of the plate 36 are an image and a HIP (Hardware Image Processing) unit 38, which specifically includes exclusive hardware for processing the images produced by the scanner 34. Therefore, this paper standard applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) A7 B7 V. Description of the invention (22) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs

Knowingly, although HIP is preferably made with exclusive hardware, HIP is also made with DSP environment or in a computer station with a general purpose:. As discussed, — 飒 仃 之 疋 制 软体… (as shown in Figure RA image processor ”typically produces the following output, explained in more detail below: a. Binary CEL b. Type characteristics c. Color CEL d. Color Defects ... e • A snapshot of the RGB image surrounding the selected defect. A downstream of the HIP unit 38 is a SIp (Software Image Processing) unit. This unit receives the above 5 inputs and performs image processing by the HIP unit 38, as described in more detail below. It is recognized that although the SIp is preferably made in software, " can be produced in at least a sub-programmable environment such as Dsp and similar environments, which typically includes most images and defect reports (equivalent to the one in Figure 1A) Defect report 3), which can be displayed to the user on the user workstation 32. Typically, the system of FIG. 1B operates in at least 3 scanning schemes or modes. The following describes the learning mode and design rules in more detail. Mode and check mode. The "learning" mode typically includes two sub-programs: ,, learning mix, and :, * A / test is better prepared for offline review before processing panel batches · ¥ κ is executed. In the learning mode, the information about the panel viewed in the design rules and / or the viewing mode is the input of the system or the "learning" function is used as a reference for subsequent processing. As shown in 7 ^, two channels Input data (quick-action data input channel and CEL _--- ^ -------- install --- (Please read the phonetic on the back? Matters before ^ | ^ this page} Ί ^ τ. -Line · -25 A7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs π 5. Invention Description (23) The data input channel is typically received by SIP 40. Each channel of the two data input channels preferably includes two sub-channels: (i ) Quickly report the sub-channel, which means that 疋 is the π report fast-moving ", and the color defect report sub-channel, which is designated as π report σ Φ color deficiency, and (丨 丨) CEL report sub-channel, and thus Designated as "Report CEln and the Defect Reporting Sub-channel, thus designated as" Report Defect, ". These input materials are generally supplied by HIP 38. The preferred embodiment of this channel is in the applicant's above-mentioned reference. It is stated in the pending Israeli patent application. It is better to demultiplex (unit 12) to obtain 3 raw data arrays, which will be explained in more detail below: type characteristics, hardware defects, and color CEL. Generally, the channel works as a converter and splitter unit 80 and 120 operates as demultiplexed raw data to separate the input poor material used by the sip 40. The demultiplexed data is thus designated as "SIp raw data." The sip raw data typically includes some information arrays. Each array It is typically stored in a data structure, and is thus defined as, a data structure array, and described in detail below. The following abbreviations are used to represent the demultiplexed data source shown in a simplified form in FIG. 1B and the drawings: C_0 C-1, C-2 ····················································································································································································································································· Each number associates a data source with the -camera in 3 cameras 0, 1 and 2 of the preferred embodiment, which data source is obtained by each camera. HD —0, HD_1, HD_2: Represents the types related to hardware defects such as nicked protrusions " Data structure array defects, " The source of the type, for this reason, the unevenness found in the binary CEL. Each number refers to the data source and one of the cameras 3, 0, i, and 2 of the preferred embodiment (please read the precautions on the back first, and then te this page). --Line · -26- 484104 A7

Read 蔼. 1 of the back

I

I

k

λ 484104 A7

V. One point in the description of the invention (25) and the PIM service is a PIM aggregation, which is used to define the position in the front view image to determine the correlation between the front view point and the appropriate view rule and method application for that particular position. PIM masks are areas that are traced by the user or created during the review phase as described below, as well as the order in which special views are performed. As shown in FIG. 1B, one of the inputs of SIP 40 is user-defined construction data 180, which typically regulates a specific SIP scheme, such as one of learning, design rules, or view mode schemes. Implementation scheme of BGA inspection with SIP The SIP is a multi-purpose flexible image processing unit, which can be easily adjusted to the inspection needs of various objects' and operates in the text of various inspection schemes. For the purpose of explanation ', under various inspection schemes, the SIp can come to the inspection of the BGA panel 36 to 6 and explain the operation. Compared to B G A viewing, the preferred scanning scheme is represented graphically as an execution diagram, and is not intended to be limited to B G A viewing, but rather means that suitable objects can be viewed using the SIP. Each of the better scanning schemes or modes operates to perform different viewing functions such as the following: 1. Design rule scheme, where the panel refers to a set of general design rule views. The contempt in this design rule scheme is intended to provide a quick look at the hardware defect and the minimum line width of the conductor width on the BGA and the small space between the conductors on the BGA (notches or protrusions). The implementation of the design rule scheme will be explained with reference to Figs. 2 • Study plan, in which the content and structure of the panel design are learned to generate references that are used during the review of similar panel batches. Learning Solution Package-28- This paper size applies to China National Standard (CNS) A4 (210 X 297 public love) — ^ ----; --------— (Please read the precautions on the back before (This page) · -Clothing Production for Employee Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 484104 A7 B7

V. Description of the invention (26) The general properties of the panel with the learning reference panel, such as the section along the conductor ------- ^ -------- install --- (Please read the back Note on the page width), and the position of the area of special interest in the panel, as described above with reference to tune 1A. The implementation of the learning plan will be described in more detail with reference to Figure 4-10. 3. Inspection scheme, where panels are inspected to determine defects with reference to a gold panel with known required attributes. Inspection schemes are typically used to inspect BGA batches used for manufacturing and to handle defects. The implementation of the inspection scheme will be described in more detail with reference to FIGS. 11-16. It is tolerated that each SIP execution diagram is connected to at least one of the data input channels 50 and 70 in FIG. 1b. The input of channel data generated by channels 50 and 70 is dynamic and tailored to the particular needs of each viewing scheme. The data in each of the channels 50 and 70 can preferably be provided in a file format for offline inspection or transmission of a Control Protocol / Internet Protocol (TCP / IP) format. Construction variables are typically provided before executing the viewing scheme to determine whether the data for input channels 50 and 70 is generated by the file channel or the TCP / IP channel. -Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. The scanning scheme is generally described in this article of the SIP construction file, where each file defines the aggregation of the review functions performed in the text of the relevant scheme. When applicable, the name of the construction file containing the execution path is specified accordingly. The following collection of preferred scanning schemes is now explained in more detail: A preferred SIP scheme, named " Design Rules "(with a build file, denoted as dr.config) will now be described with reference to Figure 2. This scanning scheme It is intended to provide a quick view of hardware defects and the minimum line width and minimum space of the conductor width on A, and can be implemented online by a file or by the user before the implementation of the solution. The Intellectual Property Bureau employee consumer cooperative prints A7 B7 Invention Description (27). This solution is typically implemented on the panel, and the use of Sightma during the learning case will be explained in more detail below. Typically, this design rule scheme loves i θ ”Manmu requires a short set time, the quality of the provided image and viewing panel is sufficient to accept people, so that the special panel can be used as the original rough indication of the ^ panel, from this-a comparative reference can be established for ^ see In the scheme, this scheme can better identify all the defects of the hardware (notches, protrusions, CEL on the rack, etc.) and all colored surface defects. The minimum line width violation on the conductor area in the BGA panel and the 2 small space violations between the conductors in the bga panel and all reports of this defect are output. The previously mentioned ㈣ is preferably identified without being associated with any external reference It is easy to recognize that if the BGA panel shows the main advantages of previous defects, it is not suitable as a reference. However, when the BGA panel passes all the above tests, it is not necessary to determine that the panel can be used as a suitable reference, and it is better to check to determine All features exist, and the features are properly located on the panel according to design specifications. As shown in Figure 2, the user-defined design rule construction data is sip 40 input. This data typically includes the minimum distance required between the conductors And the minimum line width and thickness of the conductor. The input image data related to the panel 36 is received by Shaanxi Channel 50 and CEL Channel 70, and demultiplexed into the following image-poor data: color defect data for each camera and hardware defects (engraved Marks and protrusions). The suspected defects in the input of the image data of each camera 0-2 are identified by the defect identifier 2 1 0 The identification and appropriate inspection methods are the identification of each suspected defect, the window containing the image data of the suspected defect area in the packaged soil, and the data structure. -30- This paper size applies the Chinese National Standard (CNS) A4 specification (21〇χ 297 mm)

V. Description of the invention (28) 220 operations, this window operates as a method of assembling data and is placed in a 6-becque column for processing. The defect report generator 230 (analysis H 15 in FIG. 1A) is inspected according to the identification method. If an actual defect is found, the defect S ^ %% calls for a β defect report to the defect report 24. Referring to FIG. 3, this figure is a sub-execution diagram explaining the image data flow in and between the defect identifier 2U) and the data structure storage window 2220 in FIG. As shown, the binary CEL generated by the channel 70 is provided to the defect detector 250, which operates to detect defects that are too spatially related to too much conductor and interface conductor. The output of the "too-narrow" defect is provided to the width defect report 252, and the report 252 is provided to the merge program unit 254. In addition to the report .252, the merge program table receives the color surface defect report from channel 50, and from channel 70. Receive a hardware defect report and input data 2 5. 6. The merge program unit 254 merges the report, filters the reports found in the mask field 258, and outputs a unified data report 260 in the form of a queue of defect data, which is used for Downstream processing. Masking field 258 is generated through the user interface. Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs. Once the BG Α panel is designed to be viewed as a design rule and it is determined to be suitable as a reference for possible comparison, it needs to be implemented The learning scheme of the review, the special structure of the panel is learning and stored in memory. The information about the learning panel will be used later in the review scheme, which will be described below with reference to Figures 11-16 as a reference for possible comparison of the review panel To determine the existence of defects. The learning scheme is now explained with reference to Figure 4- 10. Figure 4. This figure shows the simplified function of the learning scheme selected by the user. -31-This paper size is applicable to the Chinese National Standard (CNS) A4 (210 X 297 mm) ^ 84104 A7 B7 V. Description of the invention (29) The block diagram is printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. The user-defined learning construction data 18q is input, preferably through a general user interface 17 (Figure 1B). Panel 36 is for scanning and generating shadows. The following image data generated by each camera are used: color coffee materials 130, 134, and 138; binary CEL materials, 144, and 148: hardware (notches and protrusions) defective materials 150, 154, and 158, and models State feature data 160, 164, and 168. A learning is performed based on the image data arriving from the separate cameras, and the execution diagram is applied by applying 3 in this order: &quot; learning mixing &quot;, which is special in that it is included in the learnmixedc The nfig construction file is described with reference to Figs. 5-8; "Learning reference" is particularly included in a construction file called learn ref c0nfig 'with reference to Figs. 9-10; At the end of the plan Program specific information is received and output to a storage scheme used during the review. Typically, each scanning scheme receives raw data from the current FILE channel. The original data is retrieved from the DMA-type data input channel by a hook program initiated by the application. The program reads the information from the DMA type channel and writes the raw material into the file. The learning program then reads information from those files. The preferred embodiments of each learning execution diagram (or scheme) will now be described in detail. Reference is now made to FIG. 5, which shows the implementation of the learning hybrid scheme in FIG. The construction of the distance learning hybrid scheme is preferably stored in a construction file. In the learning blending scheme, a window is defined around the features showing the features of the mask area. The implementation diagram of this learning blending scheme is typically used by the user -32- This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210x 297 mm) (Please read the precautions on the back before @this page) Binding-. , Line 484104 A7 B7 V. Description of the invention (30) Each window of the meaning. The input file of the PIM defect filter 3 i 0 is therefore a user-friendly window. The PIM defect filter 31 is output to the defect handler 3 20, which operates to generate a view of interest expressed in the form of a SIp defect reference and a defect window reference window 330, and then uses 0 in the viewing scheme. A user-defined window can Defined with reference to the BgA type view in this learning blending scheme, including the following type areas:-Target-is an area with undefined shape features, the user desires to ensure that the particular shape is maintained in the BGa of the view;-Ball-is used to Fields where many circular pads are characteristic on BGA;-Welding fields-areas where many pads or power lines are characteristic on BGA;-Caverns-areas where there is no need for any CEL type feature;-Wafer field -The area where the chip is located on the BGA. The type of function performed by each window type is defined in a definition building case. The execution diagram of Figure 5 receives the following demultiplexed image data: a. Color cels generated by each camera 0-2, with reference numbers of 130, 134, and 138; printed and printed by employees ’cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairsb. Binary cels generated by each camera 0-2, reference numbers are 140, 144, and 148; c. Hardware defects generated by each camera 0-2, reference numbers are 150, 154, and 158; d. By each Type characteristics produced by cameras 0-2 respectively, reference numbers are 160, 164, and 168 0 -33. This paper scales Zhong Jiaxian (CNS) A4 specifications (21Gx 297 public love 1 ------ -484104 A7 B7 V. Description of the invention (31 Solution: supply of materials to the execution unit "〇, this unit and each level and the meaning of the following execution sequence. There is a reference to Figure 1-Figure 6 is a better execution Figure, explaining the execution order executed in each execution device 340. The execution diagram of Fig. 6 is related to each execution and processes the raw data from the camera. The shrinking is called the window 344 defined by the envoy to encapsulate the original The image information and data are sent to the viewing window. The 4 views are generally used at the workstation 32 (Figure ιβ). In the user interface and sent to the job reducer 342 as an event service 345. Each viewing window includes the original image information data and a user-defined window 344. The test function is different. This window is used by the job reducer. 342 is placed in the queue of window 346 and is taken out of the queue by the test manager 348. The test manager pops up a window from the queue, performs functions related to this window, and is based on the forward destination of the window. Put the window into one of the output queues. If the test manager 348 is able to complete the test on the window, the window is forwarded to the unified reference queue 35, and later in the SIP defect The processor 32 is integrated in the processor (Figure 5). If the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs prints that the test manager 348 cannot complete the test on the window based on the original image information received by a single camera, The window is placed in the undecomposed &gt; [Ninglie 3 60. Now turning to Figure 5, you can see the undecomposed queue 360 of each execution unit 34, in the proper order It is placed in the multi-cutting manager 37. The manager operates to combine each unexploded window with a suitable window to generate a combined view, so that the review can be performed. The multi-cutting manager allocates information according to the following order to Provided in Figure 8-34- This paper size is applicable to China National Standard (CNTS) A4 specification (210 x 297 mm) 484104 A7 B7 V. Description of the invention (32) At least the function specified in detail. From: =: j <Corresponding to the unexploded window, all the basic image data are ignored, and the appropriate function is based on the unified consumer window printed by the Intellectual Property Bureau Employee Consumer Cooperative of the Ministry of Economic Affairs. If the input image information h 』^ ^ ^ ^ 枓 spheroidal information, the information is provided to the spherical reference list 3 § 〇. Enter, ，,, 80 In a suitable order, the spherical measurement unit 3 8 2 receives the data and measures: head 1] especially, the spherical shape of J in the £ domain and the output SIp spherical measurement Data 3 84 to data Lv'an:, times ,, Pui Cun Sumida, here the data indicates a slight spherical measurement. If the input image data includes welding field data, the data is provided to the welding field reference ㈣39G. In the proper order, the mask extractor receives the data and captures the data masks from the different viewing areas in the window, such as in A 4 people, Japan, and Japan.丄 GA is based on the welding wire and power line in this article, as well as the reference line for aggregation and stability feature reference 394. If the input data does not include spherical shapes or mask capture is not required, the information is provided to the unified queue 398 and any masks for spherical measurement and capture. A unified queue is provided to the defect handler 320 for additional processing. It is easy to recognize that the function and sub-execution diagram of the test manager 348 (Fig. 6) is substantially similar to the original data related to a single camera performed on the combination of cuts received by the multi-cut manager 370 relative to the cuts received by the majority of camera inputs Its function. The functions performed by the viewing windows in the child learning program are defined in the file. The table in Figures 7A-7B lists the window types and their corresponding test functions (or a list of functions when the window is attached to a composite function), and provides a short description of the function. The table also describes the action and type of the function on this window and is applied to ------- 7 ------- install --- ** (Please read the precautions on the back before 0 copies (Page) Order- · Line- -35- K Paper Size Applicable to Chinese National Standard (CNS) A4 Specification (21〇X 297 Public Love) 484104 A7

Rotating state, which includes the destination A string that determines the output queue to which the window is sent. The unified queues 350 and 398 are passed to the SIP defect processor which writes defects and generates top-down references. This top-down reference is the ultimate reference for the spatial domain of interest during the review of the solution, such as the reference to Goming in Figure A. The function attached to each top-down window is performed at each spatial position related to the position-based 2 top-down window. The following functions can be performed before the top-down reference is established: a. The strip ball maintains the circular data structure only in the top-down window to define the space area of interest around each circular data structure. -b. Guidelines for the secondary window of the bar-shaped pressure welding, ordering the work to be written in different reference windows of each window, and this window is related to the characteristics of individual pads in the window. c. The bar is worn in the top-down window and does nothing. d. nop. Do nothing. · 凊 Note that during the learning phase, the Shao sub-panel is known from the inspection point of view before the time of special interest, for example, the spherical field, the chelate field and the target field in the BGA inspection application can be identified, so that each part can function Triggers for additional views in this view scheme described below. ‘At the time when the job reducer 3 4 2 was created, the data stored in each window was the data that needed to model the window part and the various viewing functions that needed to view the contents of the window in a sound viewing method. For example, in the BGA view, for spheres, one sphere in the list contains the center point and the radius are stored. For the target area, the compression that describes the boundary of the target is much -36- This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm)! LilMI! -I A Please read the precautions on the back before this page). Order ---. Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 A7 B7 V. Description of Invention (34) The edges are typically stored. = Fang ;; For the best implementation of the learning scanning scheme ^ ^ ^ 301 t &quot; works as the definition of the following attributes of the BGA test data. In the Department of Vision, the material's slight width and the conductor between the job Space-packed special materials / · 'A set of morphological features in the image of the reference object and special desires to be reduced are determined to be stable in this learning plan; C. Coffee data, which is the data of the CEL position in the reference image; 2. Feature data, the feature type and position information in the horse reference image. The execution diagram of Figure 9 uses the following input data: ”Raw multiplexed raw data of L-type features; b. Raw multiplexed raw data of CEL; c. Aggregate line reference data developed in the study of hybrid scanning schemes, line diagram 5 Explanation; / De-Broadshade Beshaw 'is defined as a masked area, in which the line width calculation is not performed, for example, in the BGA view, &quot; target π ,,, spherical, and, masking area; e. Stable feature reference, where the defined feature is regarded as a stable area. The execution diagram of Figure 9 produces the following input learning program output ... a. The square around the arena 'line width on this square cannot be calculated. The information It is stored in the learning line width unqualified window file; b. Line width reference, in the BGA view, the storage conductor width is between bga -37- this paper size _ towel specification ⑵ Q χ 297 male]) 4841U4 A7 V. Invention Note (35) the space between the conductors; C • the reference feature registered in the reference feature file; d. The original reference of the melon and feature in the CEL and type feature building case, each asset is indexed and includes feature information And generate related references As shown in the execution diagram of FIG. 9, the binary CEL demultiplexed raw data generated by each of the cameras 1440, 144, and 148 and each camera 16O, 164, and 168 are generated. The type features are provided to the ⑽ compound work unit Each compound working unit is completely related to the camera. The implementation diagram of the compound working unit 400 will be described with reference to FIG. 10. In addition, the original data of the multiplexing of the type features, although only the information generated by one of the cameras It can be used. Preferably, the information of each camera is supplied to the work learning book unit 420, which receives the stable feature data 422, which has a known and stable position on the panel 36. The stable feature data is in the learning hybrid scheme Zibai 5 Wang Shudan Fan 420 operates as a stable feature type and location on the learning panel 36 (Figure 1B), and generates a reference file 424, which is used in this viewing scheme to determine excess and missing features and The viewing panel is registered as a reference for the reference panel. In addition, CEL's demultiplexed raw data, although only the information generated by one of the cameras can be used. It is used, and preferably the information of each camera is provided to the work learning line width unit 43, which receives the aggregated line data disabling mask data 432 generated in the learning blending scheme. The aggregated line data disabling mask data The material 432 is placed in the PIM, and the points in the mapping table, MIM ,, and map • ^ τιί — — · · II (please read the precautions on the back before this page) Цιτ. 丨 Line. Intellectual Property Bureau of the Ministry of Economic Affairs Printed by employee consumer cooperatives • 38- 484104 A7 ---------- B7__ V. Description of the invention (36) Some things in the table, the line width and the area on the panel 36 are involved in this Examined in the learning reference scenario. The learning line width unit 430 operates as a learning line width, correlates the line width with the position on the panel 36 (FIG. 1B), and generates a reference file 434, which is used in this viewing scheme to determine the line width measurement for a given calculation. A reference within a predetermined range of acceptable measurements. Reference is now made to FIG. 10, which is a sub-execution diagram of the work performed in the compound work unit 400. The execution map is related to the work performed based on the original image data information of each camera. '' This execution diagram includes a geometric reference work unit 45, which is based on the input of the multiplexed original image data based on the cel and type complaint features to calculate the geometric reference to identify the CEL and type features, and to position them with the reference image. Correlate and generate output reference files. The first file is the CEL type and location reference 4 5 2 and the first wall case is the type feature type and location reference 4 5 4. Reference is now made to Figures 1 1 to 16 which show an implementation diagram of the viewing scheme and illustrate the viewing tests applied thereby. The following technique # 我 is used in the continuation description, and is helpful to understand 1 1 · 1 6. In the implementation diagram of Fig. 11, the following data sources are generated: R-1: Registration conversion between the image cutting and online image related to the referenced central camera; E-1, IM-1: respectively To obtain the excess and illegal matches of the central camera's type features. Excess match is a type feature that exists in the online image but not in the reference image. Illegal match is the type feature existing in the online image, but its coordinates match the different type feature in the reference image. -39- This paper size is in accordance with China National Standard (CNS) A4 (210 x 297 mm) |: -------- install --- (Please read the precautions on the back before this page). -丨 line. Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs π A7 B7 V. Description of the invention (37-Eighth xx = 0, 1, 2: Conversion from camera XX to the reference coordinate system, Interpreted by Rj; n 1% without change: converted by the reference calibration block mm calibration wire system, interpreted by R-1. J Dan refers to Figure U, which is used to register and determine excess and missing features.

Fig. 11 The execution diagram of Fig. 11 is executed during the inspection scheme in the registration generator of Fig. ^ A. A This is performed 1M to the original image data, including color data ⑽, =, and 138; binary data 142, _ΐ 48, hard materials 150, 154, and 158; type characteristic data 16, 164, and 168, and the scanner % Color defect produced by each of the cameras (Figure ib). The registered work unit in the specific embodiment explained by the child is for simple purpose and operates as a feature of the registered camera i, which is a central camera. The workbook 5 1G receives features 164 and reference features 520 ′ from the memory, and operates to generate registration conversions for the relative camera ι, as compared to the “excessive type feature report 532 for i and the illegal match report for relative camera 5” 4. Differentiating &lt; The conversion is split by the conversion splitter into 3 cameras to reference calibration conversions T_0, T-1 and T-2, each camera pair is a conversion, and the conversion between two calibration reference coordinate systems τ- For Α2Α, it is a conversion between reference data and online data generated by the front view panel. All data sources, original image data and calculation data of relative registration conversion, surplus and missing features, and illegal matching are the inputs of the inspection work unit 55. The execution diagrams for this are shown in Figure 12 in detail. -40- Bookmarking Paper National Standard (CNS) A4 Specification (210 ^ 7 ^ A7 Description of Invention (38) Receipt: Figure 12, this figure is an execution diagram of the viewing unit 550. The viewing unit receives the input from Ma Zheyu. Each Zhaogan compares the camera and the camera to one input, as shown in Figure U. "View Day". The inspection work unit 55 includes the first composite inspection worksheet 'based on the camera 0 and 2 to provide cutting of the outer board image The operation is based on the two viewing work units 6GG, which are cut according to the central image provided by the camera 1. See, refer to FIG. 13, which is a diagram showing the external panel composite viewing work unit 560 <function execution diagram. Each external panel composite viewing work The single A 560 includes one job = 钿 正 早早 ，, which is equivalent to the work reducer / interest window generation state of Figure 丨. It works to receive input as described with reference to Figure 12 and refer to it from top to bottom. Bottom references include, for example, the space-of-interest fields generated by the space-of-interest field identifier during the illustrated faculty scheme. Triggered on each view, such as a top-down window or a hardware defect or a type feature And not in office Triggering from the top-down window requires additional viewing. The work reducer 570 operates to generate a viewing window that identifies a viewing method and contains the required image data to execute the method. The viewing window is placed into The view window queue 572. The test manager 58 operates to take out the view window from the view window queue 572 and apply the test method to the image data in the window. It is easy to recognize that the test manager is based on data related to a single camera Therefore, the features, areas, and top-down windows extending beyond the camera's boundary or the overlay area overlapped by at least one camera are indecomposable and are placed in the undecomposed Suining column 582, which is redundant. Fully decomposed defects are placed in the defect row 5 8 4. Parallel to the operation of the work reducer 570 and the test manager 580 are -41-This paper size applies to China National Standard (CNS) A4 (210 X 297) (Mm) --- ^ --- τ ------- install --- (Please read the precautions on the back before this page)-i-line · Printed by the Employees ’Cooperative of Intellectual Property Bureau of the Ministry of Economic Affairs 484104 A7

--- ^ --- I .-------- Installation --- (Please read the precautions on the back first, then ^ this page) Two additional function test units, CEL monitor unit 590 and side defect monitoring Stand-by Fan 595 operates in compound job viewing unit 56. The CEL monitor unit 590 operates to receive a binary CEL and convert inputs and define a PIM CEL reference that will have no CEL area. The CEL monitor unit 590 looks at the CEL to ensure that no CEL is located in the area, which has been designated by the piM CEL reference as no CEL. If CEL is found in the area designated as no cel ', a defect report is placed in the defect queue 5 84. The side defect monitor unit 595 operates as a backup for hardware defects and color defects. PIM Main defines the area of the mask. If color defects or hardware defects are located in the area defined by PIM Main, no defect report will be issued. If the color defect or the hardware defect is outside the masked area, a defect report is placed in the defect part 5 8 4. . Line. It must be noted that the hardware defect source typically provides input to the side defect monitoring 7G 595 and work reducer 57 and switch p. Typically the switch P is closed, so the data generated by the hardware defect data source will not reach the side defect monitor unit 595, so the defect report is only performed by the work reducer 57 and the test manager 58 after the hardware defect has been run. Further inspections of the facilities provided were not accepted. If the switch p is turned on, a repeated report is issued, which is helpful for the test manager 58 to control and test the effectiveness of filtering real and false defects. Reference is now made to FIG. 14, which is an execution diagram showing the function of the inner panel composite work view of the early 600, compared to the central camera, the camera! The resulting image information is provided. The operation of the inner panel compound work inspection unit 600 is similar to the operation of the outer panel compound work inspection unit 560. However, in order to process the relative drawings, the paper standard is translated to the Chinese Standard (CNS) A4 specification. 0x -42- 484104 A7 B7 V. Operation of additional information of line width defect arrival shown in the description of the invention (40). Line width inspection work unit 6 〖〇 The supply is counted as a few thousand dollars; the line width calculator in Figure 1 A, provided

II ;: The line width of the image on the panel 36 (Figure 1B). The line width inspection work ... 10 operates to receive the line width reference material 612, and in the illustrated specific example, the Han application: Shanren H book data knows-= the application of the camera 1 of the central camera is due to the lack of material, As shown. Line-width inspection work unit 610: Come: Second, enter the crazy book to the binary CEL data, and the result. If a line width defect is to be detected, a line width defect report will be issued 6 1 4 〇 = Reducer unit type, which is equivalent to the work reducer / feel "fun window generator in Figure 1Af, which operates to receive input, such as Ming , And a top-to-bottom reference. Be sure to take a look at the "^ 4 Yishou / Wang Sidi" In addition to the input shown in Figure 11 (work reducer single it 57〇, and receiving his input, called Line width defect report 614, central camera excess omission report U, and central camera illegal matching defect report IM i. The printout of the employee consumer cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs, for each view trigger, such as a top-down window, Features of hardware defects or type complaints or line width defects or excess / missing or illegal matching and triggers that do not occur inside any top-down window require additional inspection. The task is to perform 570 operations to generate identification views. The view window of the method and contains the required image data to execute the method. The view window is placed in the view window queue 572. The test manager 580 operates to obtain the view view from the view window row 572. And the application of the test method to the image data in the window. It is easy to recognize that the test manager operates based on the data relative to the single camera. Therefore, the characteristics, defects, and extensions to the camera boundary 43 This paper standard is applicable to the country of China Standard (CNS) A4 specification (210 x 297 mm) 484104 A7 B7 V. Description of the invention (41) A top-down window of a foreign camera or a coverage area overlapped by at least one camera is considered indecomposable and is It is placed in the undecomposed queue 582, and the fully decomposed defect is placed in the defective queue 5g 4. The CEL monitor and the defect monitor are parallel to the operation of the work reducer 570 and the test manager 58. Unit 62 is provided. The monitor unit operates with line width defects, color defects, hardware defects, excess / missing defects, and illegal matching defects. PIM Main defines the area of the mask and the conversion is provided in 1 book 4 central camera Tj Information. If line width defect, color defect

, Hardware defects, excess / missing defects, or illegal matching defects are located by piM

When Mam is the masked area of righteousness, a defect report is placed in the defect queue 584. It must be noted that hardware defect resources, excess missing data sources E i, illegal matching data sources IMj, and line width data sources 614 each provide input to the monitor 620 and the work reducer 57 and include, &quot;. Typically, the switch P is closed. Therefore, the data generated by the above source will not reach the monitoring port. When the 620 field is closed, the defect report is only provided by the work reducer 570 and the test manager 58. It was accepted after further inspection of &lt; guilty &lt;. If the switch p is on, a double report is issued. This report is beneficial to the control and test of the test manager's effectiveness in filtering the actual and false defects. Secondly, the viewing window generated by the work reducer 570 may be considered. I think one of the two types of windows: ⑴ The pre-defined (top-down) windows obtained from the case of 'Human Reference Building'; on this paper scale, "Install --- ψ i (Please read the notes on the back first (Re-page 9) Order ·-· Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs π A7 ------_ B7 V. Description of the invention (42) ~ (2) are not in the top-down window Defects (hardware, excess / missing, line width) and features trigger surrounding windows. In a preferred embodiment of the present invention, top-down windows provide features that have been established in sound mode. For example, in a leg view The sound model has been established for spheres, welds, targets, and cavities. Preferably, the model is dynamic and capable of intelligently identifying shapes that may not accurately fit into a predefined pattern, but based on generalized inspection The formula varies in various ways. Preferably, the viewing window is provided for features that have an unknown shape or a shape that is not pre-defined according to -f. For this window, the viewing is better performed by a more general method Directly The shape in the online image of the panel being viewed is compared with the features existing in the corresponding region of the reference. The comparison mode that makes better use of the comparison of the periphery of the edge is explained below using 0. It is easy to recognize that it is related by combining the regions Localization mode function and comparison reference function, the viewing ability is generally optimized to enable the detection of objects with pre-defined features and objects without pre-defined features. ^ Hold on each top-down view window The functions of the type are defined and stored in the file. The table in Figure 15A-15C lists each window type and the corresponding test function (or a function list when the window is attached to a composite function). Figure 1 5 A- The table of 5C also describes the function of the function according to the window and the state of rotation, including the destination string that determines the output queue to which the window will be transmitted. Referring again to the execution diagram shown in FIG. The decomposition queue 582 is put into the job multi-cut manager 63. This job collects all undecomposed windows with the same identification label and the combination is selected by -45-4841 04 A7 V. Description of the invention (43) The inspection window generated by the combined cutting, executes its functions and puts the defects into the unified defect queue 640. The table in Figure 16 lists the multiple cutting management in the course of the viewing plan Possible functions of the user 630. The defect handler working unit 650 preferably operates to receive calibration conversion data for registration between the reference image and the online image in the front view panel 36 (FIG. 1B) and the unified defect report is organized by the unified defect queue 64 〇 Generated. The defect processor worksheet 7C 650 outputs a list of defects 660, which are retrieved by the unified defect queue 640. Optionally, if the defect work processor 650 outputs the file of the defect window when constructing the file guidance Accompanied by its related data such as CEL, color CEL, type characteristics, and defects in the type of response window. The file of the defect window facilitates the operation of the defect processor 650, for example. Window view printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs. ------ I .-------- Equipment --- * (Please read the caution page on the back first) Easy to recognize The ground is the SIP-based image processing hypothesis &quot; interesting, occurs, and from the image processing point of view, typically occurs in a square area of interest (R0) or only a small part of the surface area of the object under inspection Window. So a 'SIP view includes at least some view routines, which are preferably restricted to the area of interest. This approach avoids having to perform complex viewing methods on the extended image to identify viewable features that are believed to be located within the region of interest that surrounds the view-triggered view trigger. In order to speed up the calculation and permit the understanding of more complex images, preferably 3 parts of image processing are utilized. In the first stage of processing, a predefined region of interest area, such as a top-down window, is defined as an additional view trigger and is stored in offline memory. The reference image information that defines the area of interest from the top-down window includes -46- 484104 The printed clothing A7 B7 ^ _______ of the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs. 5. Description of the invention (44) The top-down window and the identification image The processing function or image processing function sequence is an area of interest that is applied to analyze the same spatial position of the top-down window in the online image of the object under review. In the second stage of processing, fast "fast and bad", preferably (but limited to) hardware-based image processing is used to identify additional viewing windows and to generate a profile of the overall objects under review. The basic image data of the online image. The basic image data of this analysis is transmitted to the Sip for SIP image processing and analysis. In the third stage of processing, 3 parts of image processing are performed better. First, calculation The task is to apply the basic image data of the analysis in general to calculate the calculation information related to the image such as line width calculation, excess features, missing features, illegal matching features, and registration conversion. Second, the window of interest is defined in the additional viewing window The surroundings are, for example, predetermined features and candidate defects output from the second stage and part of the third stage. Third, the additional viewing window is received from the top-down window created in the first stage and is an additional location outside the top-down window A view trigger is received, and an additional view window is generated. The additional view self-contained top-down window and defined in the additional view trigger The enclosing window is the basic image data of the region of interest that is dynamically packaged to include the online image of the object under review, the identification of suitable image processing functions or the image processing functions that require the order of viewing the regions of interest, and the image processing Any reference material required by the image processing function. The image processing function is preferably selected in response to the category of the area of interest to provide the best image view of the type of area in the viewing window. ^ In the construction, wisdom (and resource requirements) imaging theory ------- M ------- install --- * (Please read the precautions on the back before this page) &gt; 11 · 丨 line -47- A7 V. Description of the invention (45 B7 ,: 1 / δ 1Ϊ1 according to the special area of interest and only apply to a specific viewing window instead of the entire object ±. Point ft "See the window view trigger, R01 Or the center of the window, 々. As noted, the window view trigger may be a &quot; &lt; defect (such as excess / missing or line width reporting); t such as a type event (island in the circuit or Is the beginning "trigger twice, such as Pre-defined position defined by the user. ^, The top-down trigger can be the square R0I drawn by the user in the spherical area = the surrounding edge, or on the other hand is a pre-made two pounds. The position of the result of the learning scheme. 柃 6 various types of window triggers that will be established around the trigger ": = ,, the norm is typically attached to the middle of H q, and the trigger preferably includes the expectation in window two, R ' And the definition of the calculation performed by the data inside the Ma window. The definition is typically formulated in a special construction file. '<: Once the basic image data is visually installed, the processing function of the viewing window is based on the viewing window and the use of window internal resources \ 0 '' s processing function generates additional information attached to the window. — Some processing functions on the window are explained below. A special feature of the preferred embodiment of the present invention is the different functions that can be localized to regions of interest inside various windows. Because Λ, for example: the area of interest related to the top-down window, the smart and robust functions can be applied to the area of interest without applying these functions that are not different from the entire image. The characteristics of the top-down window pattern are special: human-48- This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 public love ------ I .------- -Installing --- c Please read the notes on the back before this page} -J ^ T. .Line. Printed by A7, Consumer Cooperatives, Intellectual Property Bureau, Ministry of Economic Affairs

Explanation of the invention (46) γ has known characteristics that can be easily expressed and viewed by appropriate methods, for example, a circle can be represented by a center point and a radius. For areas of interest related to non-top-down reference window viewing triggers, &amp; ′ comparison-based image processing functions can be better applied. For example, comparison-based image processing functions are better suited for viewing features and candidate defects with shape and size features that are unstable and cannot be easily modeled before. As a preferred method of comparison, the binary cELs are combined and converted into vector lines including directed elements during this learning phase. The directional element represents the edge between the pattern feature and the substrate in the reference image, and is stored in the memory for use during the review of the online image of the object under review. The generation of CEL and vectorized line elements is described in more detail below with reference to FIGS. 21A-21D, and the CEL-to-vector comparison function is described in more detail below with reference to FIGS. 30A-3 2B. The following best steps are helpful for understanding how windowing works. Step I: The user defines a window in a reference object around the space domain, typically including at least one unit of a predetermined pattern shape. It is expected that in the inspected object such as panel 36 (Fig. 1B), at least one shape defined by the mode is located in the spatial area of the corresponding area defined by the user in the reference object. A suitable pattern shape is preferably selected by a patterned geometric shape store that is stored offline. Preferably, each mode shape is defined mathematically. For example, a circular f'ball "882 (Figure 27) is typically found on a BGA and can be defined by the center point 884 and the radius. Step II: Define the attributes of the reference area. For example, the user can set -49- this paper The dimensions are applicable to China National Standard (CNS) A4 specifications (210 X 297 mm) -------------- install --- (Please read the precautions on the back before this page) ▲ π. --Line. Printed by A7 B7 of the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention (47) Relative to the round ball 882 position, Shi Ergan / uu ”^ Size and other characteristics of the inspection image make a small difference Parameters. Because of &amp;, the center point 884 can be defined as having to be located within an acceptable distance of the predetermined coordinates, and the radius of various positions along the circumference, ‘set me as must / zheng within a predetermined range of $ 0.5. Other extras can be set, such as whether other features, shapes, or image elements such as lines or edges are within the area surrounding the patterned shape. Step j and step Η are preferably performed in the learning phase described above. Step III: Obtain an image of the viewing object during this learning phase. Preferably, the image is obtained using the device of one of the preferred embodiments of the present invention described herein to pre-process to generate at least one form of basic image data such as CEL, color CEL, type features, and image-related Additional information. Step IV: In the online image of the viewing object, the remaining viewing areas are isolated. Each of the other viewing areas is preferably one of the areas which is spatially equivalent to the sufficient reference area in step II or on the other hand is equivalent to a predetermined viewing trigger around some form. Defects that require additional review or areas of space defined during the feature. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. It is preferred that each of the remaining viewing areas is assembled into an additional viewing view. The window includes the stored shape mode and basic image data. Perform at least one viewing function and formulate any user-defined parameters that can be required by the image processing function. For example, if the additional viewing area contains a ball 882, the window preferably contains all the CELs of the areas in the remaining viewing windows, any type features found therefrom, line information, and the like. In addition, the viewing window typically formulates the viewing function that will be based on the operation of the ball and user-defined parameters of the ball, such as -50-484104 A7. V. INTRODUCTION (48) Permissible differences in distance from required positions and different sections of the circle The permitted range of the radius. Therefore, if the viewing window is generated around the area of the candidate defect, the function of the function is to specifically identify the candidate defect, which is typically different from the function used to identify the type feature, and is included in the additional viewing window. Step V: The image processing function assembled with the additional viewing window is executed in the window based on the basic data and evaluated based on the user-defined parameters. Preferably, if the additional viewing window contains a patterned shape, the basic image data contained in the additional viewing window is processed to generate a geometric object representing the feature. What is recognized is that the basic image data is deduced from the online image of the object being viewed in the spatial position associated with the reference window. For example, if the feature being viewed in the additional viewing window is a ball, all CELs in the viewing window can be combined to produce a line representing the circumference of the ball. Once the geometry is generated, the child geometry in the viewing window is compared with the pattern in the reference window. It is recognized that many libraries of possible shapes may be built, and all possible shapes are not expected, especially those that represent defective random representations. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs Reason: 2. In a preferred embodiment of the present invention, an object that does not need to be referenced to analyze the special position of the shape of the geometric feature is provided. The geometry and expectations = are produced and compared by the representation of geometric shapes, preferably as follows: H is produced by productivity ... or hunting is known by tradition in the relevant art 51-χ29τΠΤ --- ^ --- ^ -------- Install --- (Please read the precautions on the back first and then 41 ^ this page) _Line 'The paper size is based on 484104 V. Description of the invention (49 bit map comparison. Step VI ·· The report is generated—Ma Liri does not deduct the presentation of features and attributes, the position of the sum and whether any features are defects. G IP ϋ 交 佳 二 考 Figure 17 'This picture is made by the image processor 5 (Figure 1 A) A diagram of the preferred data structure of the inputted input bets can be used in the text of the review. The original data is used as input in Figure 1, typically including the organization (word 70 stream as shown in Figure 17). Ground, this SIP channel works to convert input raw lean material into raw data array element Type representation, where the element is defined as <_: binary coffee, color melon, snap action image, type feature, defect, or color defect. The structure of the data structure array and each element type of the data structure array are described below. Detailed description. In a preferred embodiment of the present invention, the original data is received as shown in the data structure illustrated in FIG. 17 and is split and converted, for example, by the data converter and splitter unit 12 shown in FIG. 1B to be included in the data. The data structure of the structure array. The head of a typical report element printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs, explaining that one of the scanned images received by one of the scanners 34 (Figure 丨) is explained in Figure 18. As shown The header typically specifies the type of the report element, for example, the type of event that occurred in a row, how many times (cnt), and which camera or clip. Better type to identify the type of report element Design is the following 3-bit design: 〇—Report no type 1—Report color CEL defect, 2—Report defect-52- 48410 4 A7 B7 V. Invention C (50 4 CEL report consistent with the margins CEL report Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs, the number of cameras regarded as the number of cameras, for example, 3 cameras are used for imaging &quot; Hai Panel Most camera and cutting considerations are discussed below. Che Youjia ’s SIP data structure array is a data source called, data structure element. ”This data structure treats this data as a line to increase the number of“ Horse order. Each line contains a special type or element of Q or more data items (4 bytes in all training original data items). The range of accessible lines includes the smallest line in the image up to the image The largest line in. An empty lean source is identified when it encounters the following conditions: Minimum line &gt; Maximum line When the source is set up, the minimum line's 値 is set to G and the maximum line's 値 is-again 疋 -1 . The data source specifies the maximum number of lines in the scan and is preferably greater than or equal to the coordinates of the maximum lines generated by the scanner. -Therefore, each line in the scanned image is equivalent to the line specified in the data structure. The number of lines between the minimum line and the maximum line at any time is equivalent to the scanning image line prepared by the girl. Each element in the data structure array typically has only one producer and one consumer. Lines can only be added to the data source by the producer and only if the increased y-coordinate is greater than the largest line in the data source. When the-line is increased, all lines between the maximum line and the new line (excluding the new line) are added as empty lines, where each empty line contains q data. If the new line is not blank, the new line With all its elements incremented, the maximum line is updated to equal the new total number of lines. The producer of the data source must order the line [___— -53- This paper size applies to the Chinese National Standard (CNS) A4 ^ T ^; _297 Gongai_y 484104 A7 B7 V. Description of the invention (51) At the end of the scan,兮 议 # wee big line. It is recognized that: taking the largest line of :: will be equal to the maximum in the scan, and the defective line may be blank. Printed agricultural paper size 1 by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 1®Cai i ^^^ CNS) A4 Lin (⑽—mm) 484104 A7 B7 V. Description of the invention (52 Employees’ Cooperatives of the Intellectual Property Bureau of the Ministry of Economics ’print source Perform calculations. The results of the calculations are put into at least one output data source. A work is preferably independent of data disposition (for example, an internal buffer or an overflow that synchronizes between the material and the calculation), However, the work is based on certain data. The data disposal issues are typically handled by the SIP manager. The channel work is responsible for connecting SIP to the outside world when it is running on a computer, such as operating on input data to demultiplex the input data and converting The data is in a format that can be used by the SIP. Each data channel is typically attached to a physical channel (DMA, TCP / IP, or file channel). The input channel is converted to the input data as a data source. The output channel is attached to the data Source and write the source (when it is updated) to the specified channel. The use of the file channel allows β noon scans * It is easy to actually connect to the DMA or TCP / IP channel The input data can be simply injected from the building plan containing the data with the same format as the data reached by DMAT and TMP. Communication :: Obviously it is the inspection work, and calls to execute the image according to the asset department The scheduled execution map (and hence the "scanning side" is defined by the work involved in the scheme. Beiganyuan now refers to Figure 19, which is noted from the figure-the scanning scheme-which can generally be represented by this diagram. Also J The reason is in the ::: scheme, the SIP is connected to two input channels and -output channel /, -DMA channel_generating data source 2 and channel 610 to generate data source 1. With 3 computing units (work): work i uses data source 2 to generate data source 3; job 2 uses data source i instead of production to make an order and to adjust the paper size (CD 55 &gt; &lt; 297 ^ 7 V. Description of the invention (53) source ... 3 Use data source 3 to generate data source == / ιρ channel (channel ⑽ ... work 3 update :: 4: construction file execution map and the use of the channel as the formulation in the construction building case. The parameters are typically not re-created Compile or relink and run The sweeping finger case is included in the construction building case to make the expansion = software development point of view, among which-the most expensive: :: want to: include data processing and the same characteristics between data acquisition and calculation, and can be Very similar or even equivalent to Xu Xiang Αι (Automatic Optical Vision) application. Preferably, the SIP is established, so easy to work and easy to add data sources' as an inspiration for the built-in template data sources: this 'Ma increase_ data source or {data source_, a suitable type of data source or job is derived from an existing source of the corresponding basic category :, job deduction. As much work as possible is typically converted to basic Farewell (by extension of the template style). Reference is now made to Fig. 20, which is an execution diagram of the two processes. An established rule: typically handled by at least -SIP. In order to complete this work, the line graph can be divided into some execution graphs, and the data sources generated by each of the work in-can be used by another SIP :: dual purpose / if required, a data source is written by the first SIP To the TCP / IP channel and the first SIP is read by the TCP / IP channel. Minimal crosstalk results in short calculation times between processing. Installation Steps Cain Type Picture by sip line 56- This paper size is applicable to the standard (CNS) A4 (210 X 297 mm) 484104 A7 V. Description of the invention (54)

Mao Wu form material -CEL = section explains the better method ′, the information in this method is used in the sip ^ ,. The structure is not based on the basic image data, and is relative to the preferred embodiment of the present invention which enables the material to be viewed.

Binary CEL 々 In the following discussions, the following definitions of CEL (external element) are used: Lama pixel inner 疋 I single unidirectional line. Position and orientation in pixels are calculated as a function of the difference in response intensity gradients in neighboring pixels. CEL aggregation-generally defines „between two groups within a pattern formed by binary images (called images that include only two groups), and operates to define boundaries with sub-pixel accuracy. Suppose that each When CEL originates in one direction, the orientation of J σ CEL is defined, so the "white" region in a binary image that includes only white and black regions is always located to the left of the directional or directional cEL. CEL is calculated as d The function of zero crossing points of 〇G (Biance difference), as described in US Patent Nos. 5,774,572 and No. 5,774,572 and No. belonging to Caspi et al., Which is incorporated herein by reference in this manner. Other suitable edge detection techniques are described in "Digital Image Processing" by c Gonzalez and p Wintz, Addison Wesley Publishing, Reading, Massachusetts, 1987. In the present invention, the CEL is preferred The ground is calculated for the pattern edge of the red band of the RGB image. The "white" area is the area where D0g 値 is negative. Now referring to Fig. 21A, the figure explains the pixel 71 with CEL 720. 710 pixels For the pixels in the image of the object 4 (Figure 1) to be inspected, ---, 0 ---- M ------- install --- (Please read the note on the back first? Matters then 0 Page} Order. • Line-Printed by the Consumer Cooperatives, Intellectual Property Bureau, Ministry of Economic Affairs

nuz IX 84 4 A7 _B7_ 5. The invention description (55) has an angle at the origin (X, Y) and an opposite angle at (X + 1, Y + 1). CEL 720 is located in pixel 710. Pixel 710 spans a •• white ’region 730 and a π black • ′ region 740. As shown, CEL 720 is vectorized, so the white area 730 is located to the left of CEL 720. The preferred 24-bit data structure is used to represent CEL. The elements of the CEL data structure array used in SIP include the following fields: x (12 bits) is the X coordinate of CEL. Note that the lines in the image are grouped into a data array, and the y-coordinate of CEL is determined by the lines in the CEL data array containing CEL. Direction (1 bit) and edge code-(3 bits) define the orientation and direction of CEL. The final or nLn- (4-bit) defines the position of the extreme end point of the CEL along the edge of the pixel. The endpoints are points along the edge of the pixel where they intersect by a CEL. The first or &quot; Fn- (4 bits) defines the position of the first endpoint of the CEL along the edge of the pixel. Referring now to Figs. 22A-22F, the possible CEL orientations corresponding to the edge codes 0 to 5 are respectively depicted, where the direction 値 is 0 and is generally self-explanatory. Therefore, edge code 値 0 is designated as the CEL of FIG. 22A, edge code 値 1 is designated as the CEL of FIG. 22B, edge code 値 2 is designated as the CEL of FIG. 22C, and edge code 値 3 is designated as the CEL of FIG. 22D. The edge code 値 4 is designated as the CEL of FIG. 22E, and the edge code 値 5 is designated as the CEL of FIG. 22F. Referring now to FIGS. 22G and 22H, the edge code 値 6 is explained, where the direction 値 is 0. When two CELs are present in a pixel, a saddle shape occurs, as shown in Fig. 22H. In the saddle case, the independent information is expressed in a manner similar to Figure 22F. -58- This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) (Please read the precautions on the back before ^^ this Page) Equipment · Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 484104 A7 Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs π 5. Description of Invention (56). Although the edge code of 値 6 is different from the edge code of 値 5 in the later stage of combining CEL to calculate the edges representing the periphery, the edge code of 能够 6 can re-establish the saddle shape based on the evaluation of neighboring pixels. Typically, the code 値 is designated to represent the situation of each figure of FIGS. 22A-22F and 22H and the rest, the illegal “code 値” is reversed to just a skeleton event. Now referring to FIGS. 23A-23F, the equivalent of the edge codes 0 to 5 possible CEL orientations, where the direction 値 is i, and is generally self-explanatory. Figures 23G and 23H represent that when the direction 値 is i, a saddle shape exists, and is generally similar to Figures 11G and 22H, except that, The relationship between white, region, and black, and region is other than reverse. Once the edge code and direction code are defined, the orientation of the CEL relative to the pixel edge is known, and the first and final points of the CEL can also be Is defined. The first point is the point indicated by "F" in Figs. 22A-23H and the final point is the point indicated by ,, L ,,. The first column and final stop of CEL type I are preferably indicated in the loop The first point and the final point on the coordinate system. This coordinate system is the best reference in Fig. 23m. Preferably, each edge of CEL4 is divided into units represented by "a" in 4 bits. Therefore, in the example of Fig. 23 !, the first field 値 is 14 and the final field 値 is 0. Vectorized Representation of Feature Edges Referring to Figures 21B-21D, this figure explains the relationship between features and their analytical expressions as CEL and vectorized line elements. Figure 21? Shows the first and second features 750 of the plate background 760. Figure 21 (: Display includes image ^ (Please read the precautions on the back before this page) Binding: • Thread _ -59- Yang 4104 A7 ------------- V. Description of the invention (57 ) ---- --------- Install—— (Please read the precautions on the back before θ this page) Mapping table of 770 aggregation, in which CEL 78 () is seen in each pixel μ. Cel represents The sub-pixel span between the edge of the feature 750 and the substrate 76. The feature is shown as "black &amp;," because it typically has high reflectivity and exhibits negative DOG. The substrate Shown as, white,, area, because it typically has a fairly low coefficient of reflectivity and exhibits a positive D0 (}. Referring now to FIG. 21D, the figure shows a representation of a characteristic 75 ° edge (FIG. 21B), where the characteristic The 750 edge represents the edge vector 79 (). As a specific embodiment of the present invention, in the learning reference scheme described above, the CE and L of a single pixel, such as the position and orientation of CEL780, is the object 2 under review (Figure 1A ) Image calculation and record green. CEL aggregates are combined into connected elements and converted into edge vectors 790 in the same direction, and are Stored for reference in the viewing scheme explained above. The color CEL representation-line. The color CEL printed by the Employees' Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs is an extension of the binary CEL. The color CEL is different from the binary CEL. They identify and define The boundary between two different areas of the image, where each area is one of many known areas with substantially homogeneous colors. The color CEL indicates that the location of the boundary is sub-pixel accuracy, and which area is additionally indicated On each side of the CEL. The data structure of the color CEL is generally understood from the previously explained point of view of the binary CEL. Each color CEL typically includes the following fields: x (12 bits) is the X coordinate of the CEL. Please note that Each line in the image is grouped into a data array, and the y coordinate of the CEL is determined by the line in the CEL data array containing the CEL. Direction (1 bit) and edge code-(3 bits) define the orientation and direction of the CEL.- 60- ί The paper size is applicable to the Chinese standard (CNS) A4 (210 X 297) ^ 'οζ 1Χ 84 4 Printing of clothing A7 by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs The description (58) Final or ML "-(2-bit) defines the position of the final endpoint of the CEL along the edge of the pixel. The endpoint is the point along the edge of the pixel, where it is intersected by a CEL. First Or f'Fn- (2 bits) defines the position of the first end point of the CEL along the edge of the pixel. Material edge- (4 bits) defines what homogeneous color or material group exists on one side of the CEL. Note that in this preferred 24-bit data structure, resolution is sacrificed (compared to binary) when using color groups for identification. However this is a design choice and higher resolutions can be obtained with a larger number of data structures. In addition, different meanings are assigned to edge code = 7. In this case, this edge code 7 represents color bonding. Color bonding is to separate pixels of two-color CEL with edges of different materials. Representation of type features Each type feature is preferably represented by the following fields: x (12 bits) is the X coordinate of the type feature. Note that each line in the image is grouped into a data array, and the y-coordinates of the type features are formulated by the lines in the type feature data array containing the type features. Data- (20 bits) typically represents one of the following types of type characteristics: OPEN END ISLAND 3 WAY JUNCTION 4 WAY JUNCTION -61-This paper size applies to China National Standard (CNS) A4 (210 x 297 mm) (Please read the precautions on the back before β page) Binding _-· Printed by the Employees ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 484104 A7 B7 V. Description of Invention (59) 5 WAY JUNCTION BLOB JUNCTION PIN HOLE AN MINn CORNER AN n〇UT 丨, the representation of the CORNER hardware defect is preferably expressed in the following fields: X (12 bits) is the X coordinate of the defect feature. Please note that each line in the image is grouped into a data array, and the Y coordinate of the defect is determined by the line in the data array that defines the defect type of the defect. Data-(20 bits) typically represents one of the following types of defects: Small CEL defects

STRONG NICK STRONG PROT GENERAL SMALL CEL DEFECT CEL ON Skeleton-is a CEL located on a type skeleton, a defect typically formed by a line that is too thin. Excessive matching defects FEATURE- is the feature of the image of the object detected in the online viewing scheme (such as `` view ''), there is no matching feature in the reference; illegal match- the feature detected in the online viewing mode, the feature Matches the feature in the reference, but differs from the reference feature. For example, the online feature is JUNCTION, but the reference feature is OPEN END; Omission-the feature in the reference image does not have the online viewing solution -62- This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 Mm) ------ ^ -------- install --- (Please read the precautions on the back before β page) Order ·-484104 A7 Printed by the Employees ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs V. Description of the invention (60) Matching characteristics of the object image obtained. Difference between color defects Each color defect is preferably represented by the following fields: x (12 bits) is the χ coordinate of the color defect. Please note that the lines in the image are grouped into a data array. The coordinates of the color defects are determined by the lines in the color defect type data array for color defects. Color defect data (20) bits. In a preferred embodiment of the present invention, color defects are filtered by the watchdog job, such as monitor 620 in FIG. 14, and transmitted directly to the defect report. Width Defect Molybdenum 奂 The width defect report is output by line width inspection processing. The source form representation for processing detected line width defects is provided in the array of width defect data structures. The single-width defect data structure includes the following fields:, x (12-bit 疋) is the χ coordinate of the color defect. Please note that each line j in the image is a data array, and the γ coordinate of the width defect is determined by formulating the line in the width defect type data array. Category (1 bit) The width measurement category is one of the following: SPACE skeleton for space measurement; or LINE skeleton for line width measurement. Orientation-(4 yuan) 1 measurement direction, preferably formulated by one of the following: The 歹 li may be in the unknown direction, the eastward direction, and the report includes the by. Accumulation of width defects (please read the precautions on the back and then β page) Order ······· 63- in the sigmoid scale phase ’s standard (cns) a4 specification ⑵⑽1044 A7

Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs. South direction, southeast direction, southwest direction, width (15 bits). Width measurement of width defects. The other main sections of the following sections describe the other main forms of information used in SIP. Representation of fast-moving images Each fast-moving image is preferably represented by the following fields: x (8 bits το) is the χ coordinate of the fast-moving image. Note that each line in the image is grouped into a data array, and the y-coordinates of the fast-moving image are determined by the lines in the fast-moving image category data array for the fast-moving image. r- (8 bits) is the intensity of the red RGb band. g-(8 bits) is the intensity of the green r g B band. b- (8 bits) is the intensity of the blue RGB band. Line Envelope Line Envelope is the envelope data structure that surrounds the line, and is generally used in the CEL-to-vector comparison function, which will be described below with reference to Figures 30a and 32b. Now, referring to FIG. 24, the figure explains a line envelope construction 800, where &lt; Γ is a azimuthal line (envelope). The line envelope is square and is constructed to surround the line. The line envelope 800 has the same as the line 810 Orientation. The vertical distance from the line 810 to the two sides of the square is expressed by a parameter, which is expressed as the width 82, and the distance from the end of the line 810 to the envelope end is expressed by the parameter cone root width 83. Used in envelope foot (Another parameter is the angle tolerance parameter. This parameter is a parameter from CEL to vector comparison function to determine the direction of CEL is close to the direction of the line. -64-This paper size applies to China National Standard (CNS) A4 (210 X 297 ^^ ---- (Please read the precautions on the back before β page) Installation • Cable 484104 A7

Reference is now made to FIG. 25, which shows that the aggregation 850 is simply the double-line &lt; line envelope of line 810. Aggregation line 801 &lt; come set. As shown in the figure ^ ^ Installation --- (please read the precautions on the back first and then β this page) It is only the aggregation of all the envelopes that make up the aggregation line 850, the envelope of the Yonghe line 860 PIM. PIM (points in the mapping table) queries the data structure and processing points in the mapping table.在 映 # aa 泉 术, it is helpful to find the point of J in the blow table & where the special points of w + are located, and 4 are performed in the image based on the appropriate point or area in the image . PIM is preferably used by SIP filtering and monitor work.

MIM is preferably based on a simple data structure. Add additional functions for attaching different data items to different areas. Those data items can represent, for example, calculation limits. The use of MIM enables different events, such as features and defects: to be handled differently as a function of position other than the event category. -Line. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. For example, calculation and processing of areas where surface defect clusters are measured. Provided that the defect is inside a pre-defined critical field, a small member of the field can suffice to declare the position as a defect. However, if the defect is located outside a pre-defined critical field, the same fields in that field can be passed to the MIM filter so that the position is not declared a defect. By the way of the rest of the examples, it is assumed that the score is indicated as an additional view trigger. The score is characterized by a height or width exceeding a predetermined limit. In the SIP view, the height and width limits can be dynamically adjusted to the position function triggered by an additional view of the score of the object pattern under view by using MIM. It is easily recognized that the limit can be input by a user through a general user interface. -65 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 4841U4 A7 V. Description of the invention (63) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economy I Top-down Participants are top-down Reference is used to define the top-down (TD) trigger. Top-down reference to the image and category information of the area of interest around the trigger hold point for learning in the 4 learning plan and to guide the SIP to view the area of interest in a predetermined way using the viewing function related to the top-down trigger . For example, a pad-to-bottom trigger will direct the SIP to perform a set of viewing functions on the area of the space of interest in the top-down report surrounding the pad. The top-down report preferably includes the following fields: X-subpixel (4-bit): Defines the displacement of the sub-pixel in the 乂 direction from the top-down report, in which there is a top-down report definition. Y-sub-pixel (4-bit): defines the displacement of the sub-pixel in the γ direction from the top-down report. Within this pixel, a top-down report is defined. Index (24-bit): Defines the top-down report index in an array of all top-down windows of the object being viewed. Referring to FIG. 27, this figure is a simplified image interpretation of a selected element that may exist in a typical top-down view window. The representative shapes are: line segment 88, a circle "ball" 882, an indicating center point 884, And square pads, indicating open end 888. Each shape has been deduced from cel described with reference to Figs. 21A_2m. A 4 When the original data is packaged in a window, it is ready for processing. Preferably, the processing functions that operate on the window are explained below, and the data is located in the video. sip defect The SIP defect report determines the structure of a single SIP defect. sip defect report package * 66-297 ^^ · (Please read the precautions on the back before you go to this page) Install -------- Order ---- • Line! 484104 A7 B7 Printed by the Consumer Affairs Cooperative of the Intellectual Property Agency of the Ministry of Economic Affairs 5. The description of the invention (64) Contains: the coordinates of the location of the defect (X, γ). Description string. List of defects. Unified event A unified event is an event deduced by most cameras and encapsulated into a unified window. Each unified event better identifies the pixel position (X, Y). The report is based on the meaning of the location and the choice of one of the following event categories to establish the event category: top-down window feature defect width defect and each The information related to the unified event is related to the event type, and includes the above-mentioned information relative to each event. -Affine, affine conversion to a data source containing 2 for registering the camera with information about affine conversion, and registration in the viewing scheme between the object line image under view and its reference image. This affine 2D transformation is explained above. The queue of this window is the queue of sequential windows and the performance of waiting for a calculation to be performed by the task. The viewport will be used to move between viewports: to add source-queue. Windows can be placed into the queue from a job, and ---------------67-The paper size ^ secret (21Q x 297 · ^--------; --- -------- (Please read the precautions on the back and then 0 pages) Order ----line 484104 A7 B7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of invention (65) After Yuancheng's work, it is placed in another queue until it is expected that another work with this window work is out. Registration and most camera considerations are like one of the preferred embodiments of the present invention. The scanner 2 of the example and the scanner 34 of the specific embodiment in FIG. 1B use most cameras to acquire the image of the object 4 (FIG. 1A) or the panel 34 (FIG. 1B) under review. The use of most cameras is corresponding. The calibration transformation makes the input of each camera necessary. The calibration transformation is a calculated affine transformation, so each point in the panel viewed by at least one camera is mapped by the affine transformation to the same point in the calibration coordinate system. Preferably, the calibration coordinate system is set such that a single It is equal to about 5 microns. The points in this coordinate system are represented by (Xa, Ya). When the design process finds a defect, most camera scans clearly have results. The initial steps of object inspection are based on the objects 6 which are completely seen by different cameras: afterwards 'undecomposed finished products are unified-and inspected. Finally, implemented by; cameras; each finished product produced by the calibration system' Λ —The work is to uniformly calibrate the coordinate system, and the large objects represented by this part of the image are ordered and calibrated by the coordinate system. ”In the preferred embodiment of the present invention, — Formulating the geometry of the optical head: includes, for example This type is used for horsepower (preferably expressed as a diode or a wide camera like Xin Er and a camera that needs to be converted in a predetermined calibration camera ^: between the calibration conversions of adjacent camera-independent coordinate systems Heart. :: ^ —, the camera coordinated with the unified coordinate system weighs 68- reference paper size applicable in the grid (2ϋ7 公 jy I, ---- «1 ------- install --- (please first Read the notes on the back (Re-page) Order- · • Line · A7 B7 V. Description of the Invention (66) The printed clothing of the Cooperative Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs is the coordinate system for calibration. Preferably, the two-camera mode:…, camera plate &lt; 'Special review of the camera's camera firing calibration information at the time of review; and camera mode reference, when executing the f program—the relationship of the review system ^, home ..., the camera mode reference may or may not be associated with the review The photography 'mode is the same. The use of dual references can, for example, perform learning and viewing schemes on separate viewing ties. · In the present invention-the preferred viewing scheme, when the unified shadow of the object being viewed and the -reference comparison' ―Additional registration conversions are used to register the reference and online [Camera / Calibration] images, respectively. The preferred method for registration between images is described in U.S. Patent No. Harbin et al., Which is described in the 'Patent Case', which is incorporated herein by reference, and is generally adapted to the horseshoe collection And perform dynamic registration between clusters rather than through registration between individual features. Reference is now made to FIG. 26, which is a simplified, simplified image explanation of aggregates like frames 87, which can be collected, for example, on panel 36 (FIG. 1B) for use in reviewing a preferred embodiment of the present invention. The frames are equal to each other. 2 and attributes are preferably only learned in a single frame 87 °, and the overall coordinates are changed so that each frame can be viewed individually and compared with a reference. Work The following sections discuss the work performed by this SIP. The work is shown as functional blocks in the execution diagrams of FIGS. 19 and 20. ---: --- M ------- M i — (Please read the precautions on the back before this page) Order: --line · -69- 04 IX 84 4 A7 _B7_ V. Description of the invention ( 67) I / O job I / O job connects the SIP process to the outside world. Input channel works The input channel works as a FILE or TCP / IP channel. The input channel task reads the input raw data and converts the raw data into a data source. In the preferred embodiment of FIG. 1B, the input channel works in connection with the data converters and splitters 80 and 120. Output channel work Output channel work writes the data source to the specified channel FILE or TCP / IP. For example, in the execution diagram of FIG. 2, the lucky exit channel operation is related to the defect report generator 230. Compound work Compound work is the work of management and data source aggregation. When a composite job is performed, it determines the order in which the jobs will be performed. When there is a call to perform the job, the selected job is the job with the highest priority of any workable room. Measuring Line Width Refer to Figures 28A-28B, which is an explanation of the preferred method for measuring line width. Features such as the line width of the solder pads are preferably measured with reference to the type skeleton 900. The lines are defined by two lines 900 and 9 10, and each line is formed in the form of a CEL on each side of the skeleton 900. The line width is defined as the length of the line of contacting the two lines 910 and 920 perpendicular to the skeleton 900. Figure 28A shows a line width measurement of a skeleton with substantially parallel edges. Fig. 28B shows a line width measurement of a skeletal frame with non-parallel edges. -70- This paper size is applicable to China National Standard (CNS) A4 specification (210 X 297 mm) —— ^ ---- I ------- Packing—— ~ Please read the precautions on the back before B (This page)... Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs ~ --------- §1484104 A7 V. Description of the Invention (68) The line width is measured in accordance with the above design rules, learning f and review the work performed in various ways in the scheme. Preferably, the line width measurement outputs statistics of the line width and the space between all position lines of the object being viewed, and issues an error report if the minimum line width is reversed. Typically, the 'line width' is related to the line and location, while the minimum spatial resolution is the overall application. ^ Learn the line width The line width reference includes-a group of squares' and each square surrounds the image area. All line width measurements in this area have approximately the same frame. In areas where the line width has some degree of variation, it is typically impossible to establish a suitable line width reference. The learning line width operation preferably works to identify an area in which high level line width changes make it actually form a line width reference. These fields are enclosed in a square field identified as a line width that cannot be learned. View line width The view line width is the space between the measured line width and the line of the object under review. The calculated width is compared with the line width of the line in the reference image and the space between the corresponding position lines. A registration information transformation is used to correlate the individual locations of the viewing and reference images. Any mismatch between the calculated line width in the line image of the object and the line width reference specified by the line width reference at the given location, is reported as a line width defect. Registration The registration task generates a transition between the online coordinate system and the reference coordinate system of the image of the object under review. Preferably, if many cameras are used, it is converted to be provided for each camera, although it is easily recognized as -71-This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) ---- --J -------- 装 — * (Please read the precautions on the back, and then this page). Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs π 5. Description of Invention (69)

Only one camera Rong Wu # I line registration works. For example, the central camera can be used to perform reference to FIG. 29, which is an on-line coordinate system and a reference system 椤 * 结 丝 U ^^^^ pressure k system. Left-handed ^ &lt; an image explanation of the example of the online coordinate system ^ ^ 5 is an example of the on-line coordinate system and right-handed gp # 960 is an example of the rigid translation of the reference coordinate system, vermiculite and bismuth system and displacement The display only includes the rotation as rigid and-generally V;,: Ren: Yes, the registration conversion does not require any force "to include any conversions that are convergent. Study Registration-4 references to the characteristics of the registration work as a reference for registration. The coordinates (X, γ) of each sentence and the individual feature categories are better stored. ς 2 = foot meaning is "stable", meaning that they are using the inspected objects = constants within the constant small boundary, and registered as better and more knowledgeable. Foot special wants to implement. Inspector 1 master book inspection registration The data structure of the receiving input array is input, and the data structure is converted to imitate the known data structure. In addition, the data structure is output for excess defects, missing defects, and illegal matches. Refer to the way of comparison and add 'where in the study f plan before review, the reference is to believe that, the perfect object is created. The perfect object is preferably selected by scanning as a particularly safe and defective gold plate or borrowed Obtained by CAM resources. During the inspection of the square wood, the online image generated by the scanned object is compared with the perfect reference, and it is typically made to determine whether the errors caused by the online board will be referenced.

5. Description of the invention (70) The board is too different. It is typically impossible to perform a direct comparison between the reference image and the online panel 'because the position of the online panel on the scanner may be slightly different from the position where the reference image was generated. Therefore, the coordinates of a given feature on a gold panel are typically different from the coordinates of the same feature on a wire top panel. The registration work uses # 征 to calculate the registration conversion that is mapped from the online seat # system to the reference coordinate system. A brief explanation of the preferred method for BGA inspection is explained below. First, the input features are unified to a generalized feature cluster and try to overcome the inherent instability of the type features on the BGA. Identical features (such as circles) can, for example, form island features in one scan and two very close-in start features in another scan. The two-start can be defined as a single-two generalized feature, and a matching rule can be formulated to legally match this generalized feature with the island-shaped feature. The set of generalized features and the matching rules are formulated in a special construction file called a matcher table. It is also possible that the unified uniform distance defining different types of features in the construction file is transformed into features. General (please read the precautions on the back, then ®: this page)-丨 Binding-Line · The Intellectual Property Bureau of the Ministry of Economic Affairs's Consumer Cooperatives printed this registration module for better operation in order to identify that it is not in this group of online features The current reference feature is "missing", and the "reference feature" is not found in the set of reference features. The registration work also preferably identifies illegal matches. The 21 feature is detected in the online coordinate system. When measured to match a reference feature, a = method match occurs (that is, after using the registration to transform the reference feature, the cough sign is found to be close to an online character with a very different category, this 1 &amp; labor sign). Illegal A typical example of matching is between the connection and the island. -73- This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 Public Love A7 V. B7 Description of the invention (71 Excess / Omission / and Illegal matches are reported as excess / missing defects. In a preferred embodiment of the present invention, the registration work is only based on the &amp; work 'side camera feature is located in the center of the central camera or in &amp; The split registration conversion obtains the registration information and splits it into 3 =: Be ;, a crazy book information is relative to each camera. In another preferred specific: in her example, 'where the registration data is obtained for each camera, this work is It is necessary to consider that the registration work can be performed based on the calibration features. In the Shaanxi moving shell material packaging work, the input image data is processed and window 1 is created to include the complete image data of the required area. The window is known to color by cluster Defects are calculated together, so each snap-action window will contain snap-action data around a particular cluster of colored $ sags. When snap-action data is packaged into a window, it is also better to include a queue of color defects. ^ PPM is established based on information. Fortunately, in a preferred embodiment of the present invention, the window is put into a ilHIIJi selected as a fast moving image. — — — — — — 11 (Please read the precautions on the back before This page) Order! Printed by the Consumer Goods Agency of the Intellectual Property Bureau of the Ministry of Economic Affairs ^ fojianjl. The input data is processed and one of the rGB images containing the continuous fast motion field is processed. The output file is generated. The monitor ^, the monitor compares the internal data of the input data source with the specific PIM. If a predetermined type of predetermined data is found to exist in the internal prohibited area as described in p 〖M, the damaged data is converted into SIP defects .SIP defects are accumulated in the 〃sip defect data structure, which is assembled into a window, and the video

484104 A7 V. Description of Invention (72) The window and all its defects are put into the window output queue. At least the characteristics and defects are preferably monitored by a monitor .. Color defects, depressions, line width defects, and pattern features. The foot work reducer encapsulates the features, CEL, and defects for a specific window (square area) 'to form a region of interest for the remaining downstream processing. The window is preferably controlled from top-to-bottom events and width defects, excess / missing defects, and hardware defects. The top-down view of the forming part from top to bottom reference is preferably established during the learning program 'and preferably the CEL and features are converted to the online coordinate system before the top-down view is encapsulated (use registration Conversion). In order to minimize the number of windows created and also surround all the information of interest (top-down windows and windows around the defect area), a row design is typically used: all the first package data of the top-down window adds each package Window to package window row. Consider all defects and line width reports: y Order line 1 · Take the first defect that is not in the window of the first build (the order is first increase then X coordinate). 2. Create a window in the center of the selected defect, encapsulate all the data inside this window and add it to the queue of the package window. 3. If there are no more defects inside the previously created window-move to step 1. At the end of the packaging process, all materials (CEL, color CE) L, defects, -75- This paper size is suitable for BS Office (CNS) A4 specifications (2W χ tear public issued A7 -___ B7 V. Description of the invention (73 ) Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs? Π Features, width defects) are encapsulated into at least one window, and the window is output to the TV®. Please note that the window can overlap, where the The event data items are encapsulated in each overlapping window. The star data participant stands for this task to take the input original data and create a reference for the original data. The reference is suitable for quick access in the monitoring scheme. The filter is attached to the PIM It is used for filtering. We assume that the piM is sufficient in the reference weight coordinates. This work receives the input data of color CEL, features, width defects, defects and affine transmissions and considers the elements of the input data source. First This element is transformed into a reference (calibration) coordinate system using the input affine transformation. If no input transformation is formulated, the calibration transformation is used to transform the camera element Is the calibration coordinate system. Then consider each element of the input data. This element is converted into a reference calibration coordinate system and checks whether the element is inside the specified filter area of the phm. The construction file determines whether to filter this element. If the element is not filtered Time, it is put into the output data source. The construction variable typically controls whether the element should be filtered if it is inside the specified area, or whether it should be filtered when it is assumed that it is outside the specified area. Retrieving information from the window works To pop up the window from the input data queue and retrieve data from the window as needed. When completed, the window is put into the output queue. Production-production area and stability refer to this work by referring to the data. Enter the input window to write the auxiliary information, which is -76- This paper is suitable for financial standards (CNS) A4 ^ 7ir〇x 297; ^ T) (Please read the precautions on the back before this page) A7 A7

V. Description of the invention (74) Typically includes: (a) various types of PIM areas and (b) all features are declared by the previous function that operates on the window, "stable" features. Quick measurement This work is entered by input The window writes auxiliary data. The data typically includes a file containing all the information found on the circle in the window, as well as other statistical buildings including the distribution of the circles contained in the panel. Test management The input window queue pops up a window and uses the image information data contained in the window to perform the function related to the window. According to the output of the function, the following actions will be decided: ^ ^ -------- Install --- (Please read the precautions on the back first, then ^ this page) Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs, Employee Consumption Cooperative, printed π Table 1 The test manager actions corresponding to each rotation state generated by the function, the rotation code action error window was deleted And the task was deleted by mistake. The window was deleted. The task continued to pop out the next window. The forward window was transmitted to the forward queue. The forward child window was after the job was completed. Each sub-window of the window is transferred to the forward queue, and the original input window is preferably deleted. In a preferred embodiment of the present invention, a sub-window can be formed, for example, welding alone in the welding field Around the pad. Preferably, if there is no function attached to the window, the state of rotation is 0K. Multi-cutting camp rationale The multi-cutting manager works better to receive packets from different cameras. -77 -

• Line · 04 1X 84 4 A7 _B7_ V. Description of the invention (75) A window for disassembling data. Preferably, the window is located in a calibration coordinate system. The multiple cropping manager unifies all windows with the same window identification number and stores the unified window into memory. The unified window surrounding all child windows is created and the function attached to the unified window is performed later. Based on the output of the viewing function, decide what to do as follows: Table 2 Multiple rotations of the manager corresponding to each rotation state Action rotation code error 0K Forward subwindow In addition to the above, the action window is deleted and worked Rotate window by mistake is deleted. Work continues to pop up a second window The window is transferred to the forward row. For error is error ------ I! — — — —Install i — ~ Please read the precautions on the back before this page) Order-Printed by the Intellectual Property Bureau Employee Consumer Cooperatives of the Ministry of Economics, it is typical that If no function is attached to the window, the rotation state is ΌΚ. % Trap writer / The device 1 writer preferably accepts 4 file names from the build file: (a) Output window file name: is a file containing a window. A '' none '' indicates that the file name is not the correct file name and that the file is typically not generated. (b) Defective window file name: It is one of the files containing defective windows. '' None '' indicates that the file name is not the correct file name and that the file is typically not generated. (c) Defect file reference coordinates: Includes defects expressed in reference calibration coordinates. 78- This paper size applies Chinese National Standard (CNS) A4 specifications (210 X 297 mm). Line _484104 A7 V. Description of the invention (76 Ministry of Economic Affairs) Printed by the Intellectual Property Bureau employee consumer cooperative == :, n ° ne ', 値 indicates that the name of the robbery is not the correct name of the building case and this bridge case is typically not generated. RLd) Defective building case online coordinates: To include online The deficiencies of the coordinate representation: two_e &quot; 値 refers to Qian Qian κZheng Yun case and this file is typically not generated. This defect composer performs the window from the input 1 1 pop-up window and preferably as 1. If the window contains defects: If you can not create a case with a defective window or output a case, the child Visual clover is stored for later processing. Otherwise, the window removes all its data except for defects and is stored for later processing. 2 If the window contains no defects: If an output window cannot be generated, the window is stored for later processing. After scanning, f, all previously saved windows are processed as follows: All defects generated by the All View® are accumulated. In the accumulation step ', a second vector of defects is established ... 1. The first vector containing the defects represented by reference calibration coordinates. Defects in each window are converted into a reference coordinate system by using the Self-reference affine conversion, which is preferably present in each window, and transmitting the visual content to the reference calibration coordinate system. Defects are added to the vector only if they are not in the mask field defined by the plM in the input PIM data source. If no PIM is defined, all defects are added to the vector. Then at the time of the mid-line at that time, the 79-paper size _ t _ specifications ⑵ "297 public fishing 484104 A7 printed by employees of the Intellectual Property Bureau of the Ministry of Economic Affairs Consumer Cooperatives V. Invention Description (77) 2 · Including the online calibration coordinates The vector of defects. This direction f is generated by the first to f by applying the registration information formulated for the job input. Please> Wang Yi When no input affine transformation is defined, the identity transformation is used. If included When a correct output file with one of the reference coordinates is defined, the first direction I is written to the child file. If a correct output file containing one of the online coordinates is defined, then the second vector is written into the file. Each of these defects The file contains a set of lines, one line corresponding to each defect. Each line typically has the following format ... XY-Defect category-Frame index where X and γ are the coordinates of the defect (indicated by reference calibration or online calibration coordinate system); The defect category is an enumeration of defects; and the frame index is an index of the frame containing the defect. Then 'if the file with the correct defect window corresponds to the name, we will The defective window is written to the file. This situation is particularly beneficial for visualizing the defective window. Finally, if the file with the correct output window corresponds to the named file, the function attached to the window is executed first, and then by considering the The rotation of the function may determine how to proceed: 80- The paper size is suitable for K ^ Fresh (CNS) A4 regulations!) F Please read the precautions on the back before I® · Install 4 ^ 1 · -line · 04 11 84 4 A7 B7 V. Explanation of the invention (78) Rotation code error 0K Forward and forward sub-windows In addition to the above, the actions of each rotation state generated by the functions in Table 3 work to rotate in error. All windows are deleted and the windows are written to Required files. Such as the 0K case. All sub-windows in this window are written to the required files such as error cases. (Please read the precautions on the back before this page.) Printed by the Intellectual Property Bureau Employee Consumer Cooperatives It is recognized that if there is no function attached to the window, the rotation state is typically 0K. CEL for drawing calibration The CEL work for drawing calibration takes into consideration CEL. If the CEL is not in the formulation When constructing the inside of the editing area of a file, it is converted by the camera into calibration coordinates and written into a file represented in a special graphic format to make the file suitable for viewing. This task is useful for checking the calibration conversion in the viewing system. Preferably, this work is performed for all cameras and the output files are displayed on top of each other. A good calibration conversion is the conversion of the CEL from the two overlapping cameras, so in the overlapping area of the two cameras, the CEL of the first camera is accurately located Above the CEL of the second camera. The function is a package of a viewing method for the image information contained in the viewing window. The function is preferably started by the window orientation task in a predetermined order to view the image data content of the window. The task It is better to test as explained above -81-This paper size applies Chinese National Standard (CNS) A4 specification (210 X 297 mm) line · 484104 Printed by Consumer Cooperative A7 V. Description of Invention (79) The manager or Multiple Cut Manager. The startup task applies the function to the data in the window and uses the window-specific function as an argument. This function performs one of the calculations on the window and one of the three rotation codes described below. Based on the turn code, the call job decides what to do next. Each function preferably receives all parameters required for the calculation from the construction file, and captures image data information required for the calculation from the window. Preferably, when more complex calculations are needed on a given window, the functions are modular and linked together. The function is better to go back to 3 possible one of them: OK, instruct the function to continue its calculation. An error indicates that the function failed the calculation (for example, because of an error or a serious problem), or it indicates that the function continues its calculation but indicates that the remaining calculations are required to complete the calculation. This task is performed by transmitting the window forward to the forward destination and assigning additional functions to the window. The startup job typically knows how to send the window to its forward destination and how to activate the forward function. Forward subwindow, indicating that the function continues its calculations but indicates that the remaining calculations are required to complete the calculation. This task is performed by transmitting all child windows in the window forward to the forward destination and assigning additional functions to each child window. The startup task typically knows how to pass the child window to its forward destination and how to activate the forward function. No operation (NOP) function This function does nothing, and preferably rotates to 〇Κ 値. -82- This paper size is in accordance with Chinese National Standard (CNS) A4 specification (21〇x 297cm ^ --------------- I--install --- (Please read the Note on this page) Order · ·-line · 484104 V. Description of the invention (80)

Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. This function is a better function to perform one of the list functions provided in the media 4 Huoyao and Qishishi restructuring files. These functions are performed in the order specified in the build file. Forward function This function obtains the meaning of A from the construction project, 1 the name of the forward function and the forward destination for the rest of the processing. The forward direction g A,, .. ^ Month J Liyue dagger car accepts the flag of making special anomalies forward, typically including: Anomalies forward only when inside the camera, operates as a confirmation window Only forward when fully within the camera's boundary. The anomaly is forwarded only when a defect is found, and it works to make sure that the window only transmits forward if it contains a defect. The abnormality of the seven sub-windows has a false flag and the operation is in a state of rotation, so if the child flag is 眞, the state of the child window is transmitted forward. Otherwise, it will turn forward. The window is preferably not forwarded if the following conditions are encountered: 1. There are no defects in the window, but it is assumed that the window is set to flag only if the defect is present in the window; 2 The viewport is not completely inside the camera boundary, however it is assumed that the viewport is set to be flagged to move forward only when it is completely inside the camera boundary. Transform window into reference coordinate function This transform window uses reference coordinate function to remove non-convertible data items from the window and convert window coordinates from online image coordinates to reference coordinate system. The function of removing non-convertible data items from the window The function of removing non-convertible data items from the window is better to remove the non-convertible -83- This paper size applies to the SS home standard (CNS) A4 specification (210 X 297 public love) — ^ —J .------- installation · — ~ Please read the precautions on the back before β page) Order · -line · 11 84 4 A7 _B7__ V. Description of the invention (81) Convert image data, for example The CEL that exists in the window. Watch function The watch function scans all input CEL in the window. Each CEL found within the quadrilateral surrounding the reference window is reported as a defect. This feature is typically used to identify areas that lack any features. For example, in the BGA inspection article, the monitoring function can be applied to areas defined in a top-down window that must have absolutely no features. In-window defect filtering function In-window defect filtering function is attached to the PIM data source. This function receives the PIM with reference to the calibration coordinates. When executed, this function converts the input defect into a reference calibration coordinate system and checks the calibration defect with the PIM. If the defect is found inside the mask area specified by PIM, the defect is filtered out, otherwise it is put into the output list of the defect. Linked element functions Linked element functions are established by the CEL stored in the window (including binary CEL and color CEL). The resulting connected elements are placed in a window, preferably further processed by the vectorizer function, as explained below. Vectorizer function The vectorizer function obtains the connected elements and approximates each connected element by a straight line, so the difference between the connected element and the straight line is not more than the specified tolerance established by the construction file. The ball (circle) patterning and viewing functions are preferably used in the BGA viewing ball patterning function in this article. Generally speaking -84- This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) •- -—0 --- J ----------- (Please read the notes on the back first and then this page) · Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 484104 A7 V. Description of Invention (82) The description is as follows. The ball patterning function is a model of the patterning category, and is suitable for viewing online images of objects under the view of the text from the top-down window. It is a ball patterning of directly connected elements ~ The ball patterning processes the connected elements found inside the viewing window. 1 The force can operate in the learning phase as a reference preparation mode, and in the viewing stage, it determines the viewing of the online image of the object under review. The characteristics of the ball. If the connected element does not already exist, this function generates the connected element by CEL. The viewport should only contain balls. Therefore, if the connected elements in the window are closed, "assuming a single ball", but if the connected elements are open, the part of the ball is assumed. This function analyzes the connected elements to determine if they are a ball. For example, by testing (points and then measuring the radius along the periphery of the connected element to determine whether it is within a predetermined tolerance. The circle in a single connected element The 'beixun' in the state remains in the data structure represented as a generalized circle. Prepare the ball window for reference. Prepare the ball window for reference. Remove all but the ball from the window. Refer to the reference ball and the online ball. Compare the reference ball with the online ball. Compare the ball in the reference image of the object being viewed with the ball in the online image. Preferably, the position and radius of the online ball are compared with the reference information of the ball. If the ball on the line is similar to the radius and has a specified tolerance, the ball in the online image is regarded as the ball matching the reference image when the brother is within the reference and the distance is specified. CEL and vector comparison CEL and vector comparison function compares the reference vector obtained from the reference window (-85-This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 public love __ ------------- -Install --- ~ Please read the note on the back first Matters needing attention®: this page).; Line · Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 484104 A7 ---------- V. Description of Invention (84) Candidate defects or actual defects. It can be seen from the explanation that, regardless of the projection 1070, the aggregation of all CELs 1060 is located in the area defined by the envelope 1030. Therefore, as a preferred embodiment of the present invention, the CEL to vector output is in the envelope 1 〇30 The defect of the external CEL 10 60. In FIG. 30C, it is also seen that the second cluster of CEL 1080 is located entirely outside the envelope 1030. Each CEL 1080 in the second cluster is flagged as a defect. It is easy to recognize that A CEL such as CEL 1080, which is located outside the envelope 103, can be filtered by the defect handler function and reported as defective or too uninteresting. The CEL to vector function is preferably performed in the following order: A viewing window such as a window 1000 is to check the presentation of CEL and make sure that the number of CELs in the window does not exceed the maximum number of CELs allowed in the viewing window. Preferably, the CEL number is calculated by counting the midpoint of each CEL. Ground is variable And can be set or changed in response to the position of the window in the image. Window 1000 checks the presentation of the reference polyline 1 0 1 0. If the viewing window does not contain the poly 1 100, the defect is turned back. Better Yes, aggregation lines 1010 and CEL 1030, 1040, and 1060 are aggregated into micro-registrations. Now refer to Figures 3 1-32B, which shows and explains the good method of micro-registration. Among them, Figure 3 1 is suitable for micro-registration in the viewing window. A flowchart of the preferred method for registering reference images and online images. Figure 32A is a simplified image interpretation of CEL 1110 gathered without reference to a polygon (polyline) reference micro-registration. Its reference registration _-87 _ This paper size applies to China National Standard (CNS) A4 specification (210 X 297 public love) ---- ^ ------------ installed --- (' Please read the precautions on the back before β page) tr .. 丨 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 Printed by the Consumer ’s Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs? Agriculture V. Description of invention (85) Interpretation of images gathered by CEL. The micro 1 master book is preferably performed as follows: y 酽 1100 draws each CEL 111 in the viewing window 1120 to be checked, and at least _envelope n surrounding the reference aggregation line 114 (): CEL 1 1 1 〇 Identified and similarly labeled; Step 1200, · Each CEL 1110 within the envelope 1130 is the direction and check 0 The direction of each CEL 'according to the viewing requirements within the angle tolerance provided by the construction file, and the aggregation line 114〇 Same, expressed as "match" cel 1210; Step 1300: Least square meter #Performed according to the distance between each matching arc i2i〇 and its neighboring aggregation line, and the cele 2 i 〇 aggregation phase in the online image is aggregated Line 1 140 moves until 値 is minimized. When the 値 is minimized, the image is micro-registered, as seen in Figure 32β. After the micromania book, the CEL-to-vector comparison is performed as follows: The envelope 1030 (Figure 30A) is established around the reference aggregation line, with a width equal to 値 of the construction variable. Later, continued, CEL 1040 was compared with envelope 1030. If CEL 1040 does not find any envelope, a defect is reported. Preferably, each envelope records the number and length of all CEL 1040 present inside the envelope 1030. If no CEL is recorded in a predetermined envelope in the viewing window 1000, a defect indicating that the affected envelope is blank is reported at the midpoint of the envelope. The function of creating a reference window in the force field is preferably performed in the B GA view. The function of establishing the reference window from the field of pressure welding is to analyze the connection. -88 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm (please read the precautions on the back before this page). . 04 1X 84 4

V. Description of the invention (86 elements and the found connection element + county killer containing '^ m ^^ ^ ·, 疋% horsepower force line' part of the functional entity. The user ... construct variable 'is adapted to, and used to determine the virtual The processing of φ ^ Κ 伢 刎, and the structure of the mouth is in progress. When the power line is detected, Zhu Kao's calculation of 3 viewing interests is performed: / 可 -critical 'Xiao preferably exhibits strictly defined attributes such as location-㈣Γ ί has Relatively necessary attributes and positions to a large degree of latitude area-instability are used for location; areas where horse defects are not considered. Unstable areas are only viewed if requested by construction variables. The function of creating a welding reference sub-window is also better Ground operation in BGA inspection 'and suitable for detecting whether the inspection window surrounds any welds. Mind can analyze all connected elements in the window, as well as identify those connected elements of the execution part. Critical, non-critical pure In the construction of the calculation of the unstable region where there is no change, the periphery of the homogeneous color group (identified by M La) is swelled in a pattern, and it is viewed in the BGA (preferably concrete). In the embodiment, all other materials are used to separate the masked areas. To vector error classification /, land, all CEL to vector errors are considered defects. However, sometimes CEL to vector can indicate "surface defects," which appear. Surfaces The defect is an element of the external connection presented in the reference material. Now refer to FIG. 30C. Assume that the reference aggregation line 1010 is a pad. As in the preferred embodiment, if the CEL aggregation 125 is detected as having an equivalent = In the reference image, if the polymer wire of the required pad 1010 has the opposite polarity, it indicates "pinhole". The polarity is CEL or the orientation of the polymer wire, respectively. •-· ι— 丨 丨 丨 丨111 · · 丨 I (Please read the precautions on the back before this page) Order: --Line Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs -89-

^ 4104 A7 B7 V. Description of the invention (87 褒 Processing Now refer to Figure 33. This figure is as good as showing the operation of the combined image post-processing system. If it can be converted into a process chart, the display device 26 (Figure 1A) In the figure ^ work behind the image if as the image post-processing product depicted by A =: ==: =, the number of which is the only two of the key gold on the = = again,. ,, and if the needle When the number of holes exceeds the fixed limit, in this specific embodiment, the defect system in the image of the object is a small defect. The regular defect in the image is equivalent to the defect 'typically not tolerant, and a small number of small defects. It is equivalent to the regular defect seen in the object, which is typically fault-tolerant. The holes provided above can be regarded as small defects, and missing features such as missing welds can be positive defects. Step disc 3305: User-defined The definition of defects in the image: 丨 similar to those shown in Figure 34_42 allows the user to formulate the defect step 3310: if the small defects of any category is the highest tolerance, then for the small defects of each category, use Minor faults The maximum number (χ). The number (χ) can be a defect or a defect. In fact, a small amount of +1 in a particular category is not tolerant. Step 3315: The image is viewed, preferably based on the use reference At least one image processing method shown and explained in Figures 1A-3 2B of this issue. Step 3 3 2 0: Determine the tree to determine whether a defect is found. The middle hand guideline varies from Ming to Ming. (CNS) A4 specifications ⑵G χ tear public love 484104 Α7 -___________ Β7 Printing of clothing by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention (88) Step 3325: If no defect is found, the image is regarded as κ. Step 3330: The further decision tree then classifies the defect as regular or small. Step 3 3 3 5 • If a regular defect is present, a defect is issued. Step 3 3 4 0: Small defects of each category are counted. Step 3345: No defect will be issued if the number of small defect classes is less than the permitted small defect number of a given class. However, if the number of small defect classes is greater than the permitted small defect tolerance number of the class, that is η &gt; χ, the defect report turns at step 3335. It is easy to recognize that the user of the image processing system includes a small defect post-processor described with reference to FIG. 33, and typically determines the parameters and criteria for image viewing, and the same or Separation parameters can be used for the user-defined view of the area of interest, and for the identification of defects identified by a single processor. The second is that the processing unit description is not intended to be limited to the previously mentioned method. Processing methods can be used. Suitable post-processing methods can include, for example, further analysis of the function of the area of interest surrounding the candidate color defect, the function operation is to identify and classify defects formed by oxidation and photoresist residues, and can be improved by sound Additional image processing of selected sections of the image to filter other suitable defect candidates. One boat: The user interface recognizes that the image processing system of the present invention is particularly flexible, and allows most viewing parameters to be identified and controlled. The parameters include, by way of example and not limitation, edge parameters, correction, repeat, Iρ, color And light. Viewing criteria are also preferably developed by the user. This parameter can include

(Please read the precautions on the back before β page), install *-• line-484104

Examples and methods are not limitative, but can include categories, location holes, top-down views of the top-down window, and "final pads" can also be provided, such as minimum :: force :: analogue. See the remaining parameters J ^^ line. The present invention is preferably related to the Tron domain on several pieces to be inspected. The parameters are the ones of the present invention, which is generated by image processing 2 during the definition period of Yuanyuan Gold M and the person with reference to Figure 34-42. The establishment and operation of the preferred embodiment. ', Zhongzi, such as this = 34 is a labor display that allows users to formulate general inspection criteria. Figure 35 Screen display of Ma Yun comparing the criteria for pressure welding inspection by the user For example, the user can set the target display criteria for the labor display. For example, the child can display the power line inspection guidelines for the child display. Figure 38_39 is an example of the screen display that allows the user to formulate the ball viewing guidelines. Advanced image review guidelines and: a preferred embodiment of the present invention, two parts = for defining the area of interest in the object to be viewed, such as: The graphical representation of the viewed object is provided, so that the anvil will be inspected 』1 &lt; It means that the device is typical ~, (assistance, specified on the category of particular interest in the review. In the second part, the table m ... w indicates the area of non-interest. Parameters &amp; are provided for the designated special feeling of this patent Part of the document reveals that the copyrighted material is included. The -92-Pack (please read the precautions on the back! ^ This page) Order-• Line-Printed by the Employees' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs ^ 4104 2. Description of the invention (9〇) = ownership, no objection to the disclosure of patent documents or patents: reproduction's office appears in the file of the Patent and Trademark Office or records: All copyrights are reserved. The invented software component can be produced in ROM only if needed. The software component can generally be produced in hardware and / or Dsp unit using traditional technology if needed. It is recognized for clarity Various features of the present invention described in separate embodiments can also be provided in a single specific embodiment. Conversely: for the sake of simplicity, the various features of the invention described in a single specific embodiment can also be It is provided separately or in any suitable sub-combination. It will be recognized that for anyone skilled in the relevant art, the present invention is not limited to any embodiment that has been specifically shown and described above. Rather, the present invention The scope is only defined by the scope of the subsequent patent application. «— — ^ • — — ^ • — — — 1 — — · ^ · 11 ί Please read the notes on the back before this page} Printed by Consumer Cooperatives-93- This paper size applies to China National Standard (CNS) A4 (210 X 297 mm)

Claims (1)

484104 A8 B8 C8 D8 1. A method for inspecting an object, the method comprising: creating a reference image of a representative object, the reference image including a first representation of at least a portion of a vectorization of an inner boundary of the image; obtaining an image of the object under inspection, including the image Inner edge representation; and-comparing the second representation of the boundary with the first representation of at least a portion of the vectorization of the boundary, thereby identifying defects. 2. The method according to item 1 of the patent application range, wherein the comparison step uses _ user-selectable variable thresholds for the acceptable distance corresponding to the depression between the boundaries of the gates of the first and second representations. 3. A system for image processing, including a boundary recognizer that operates to produce representations of known element boundaries in the image; a hardware candidate defect identifier that is operable to identify candidate defects in the image with hardware Software candidate defect viewer, which receives the output from the hardware candidate defect recognizer and represents it in software usage boundaries to identify at least-false alarms within the output. Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 4. If the system of the scope of patent application No. 3, the boundary identifier includes a hardware boundary identifier, which operates to generate a representation of the boundaries of known elements in the image by hardware. 5. If the system of item 3 of the scope of patent application, the system also includes a software candidate defect identifier, which operates to identify other candidate defects in the image by software. -94- This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) 6. If the system of the scope of patent application No. 5 is applied, the software candidate defect viewer is also received second by the software candidate defect viewer The output and a soft-following boundary are used to identify at least one false alarm within the output. 7 'The system of claim 3, wherein the hardware candidate defect identifier uses a representation of a boundary to identify at least some candidate defects. 8. The system of claim 5, wherein the software candidate defect identifier uses the representation of the boundary to identify at least some candidate defects. 9. A system for image processing, including: ^ a learning sub-system operable to define a first and a second field of an object under review, wherein the first field each contains at least one known critical object element and The second field does not contain the known critical object element; and the defect detector's operation is related to the known critical object element: prior knowledge uses the first procedure to view the first field, and uses a different order than the brother The second procedure to view the second field. Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs, Shelley Consumers Co., Ltd. 10 · The system for applying for patents_9, where the second program includes a hardware review of the second field, which operates to identify candidate defects in the second field and includes Subsequent software inspections in the second domain only operate if the at least one candidate defect found in the second domain is found to analyze the at least one candidate defect found in the second domain to identify false alarms. U · —A method for inspecting an object, including: In the first stage of inspection: identifying the position of the at least one candidate defect; and for each candidate defect, the position in the Θ us &lt; candidate field Decided to use at least part of the peripheral area that is based on this identification step. _ -95 _ This paper size applies the China Standard (CNS) — 4104 ABCD, patent application scope. Printed by the Consumer Property Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. Review of small and shape; and in the second phase of the review: Review each candidate area to identify the defect. 12. The method of applying for the item n of the patent scope, wherein the first stage is set to 0 by hardware Jb. 13 · The method of applying for the 11th scope of the patent scope, wherein the first stage is completed by software. 14. The method of claim 丨 丨, wherein the determining step includes determining the size of a candidate field at a known location. 15. The method according to item 丨 丨 of the patent application scope, wherein the second stage includes performing different inspections according to the criteria of the candidate field or the candidate defect. 16. The method according to item 丨 丨 of the patent application scope, wherein the second stage includes performing different inspections according to the characteristics of the at least one candidate defect identified in the identification step. 17. The method according to item 11 of the patent application scope, wherein the second stage includes performing different inspections according to the function of the object part. 18. The method according to item 11 of the patent application scope, wherein the second stage includes performing different inspections according to the criticality of the function of the object part. 19. · A modular image processing system with a user-definable image analysis function for use in conjunction with a scanner; the system includes: an image processing engine that receives at least one image data application stream to be analyzed by the scanner ; And an engine builder operable to receive at least one user-definable image analysis function to be performed on the image data by the image processing engine -96- paper size 2T ^ 7—n Please read me Matter rebinding page ABCD Printed by the Consumer Property Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs VI. One of the definitions of the scope of patent application; ―The channel image data includes performing different fir image analysis functions, and according to the current definition of reaching the M-structor from the sequence definition ', not only the relevant parameters are different from each other. 20. A method for automatically optically inspecting an object, including: In a first inspection step: sufficient for most regions of interest for image processing, the salty region of interest includes at least one region of interest defined by a user At least one region of interest that is automatically defined by optically viewing the object; and the image surrounding the region of interest to the image processor; in the second viewing step: automatically processing each region of the region surrounding the region of interest The image determines the appearance of the defect in the article. 21. The method of claim 20, wherein each image of the area surrounding the region of interest is smaller than the image of the object. 22. The method according to item 20 of the patent application scope, wherein the area of interest is automatically satisfied by optically inspecting the object, including the candidate defects of the pattern formed on the object. 23. The method of claim 20 in the scope of patent application, wherein the region of interest automatically defined by optically inspecting the object includes a predetermined pattern feature of a figure formed on the object. 24. The method as claimed in item 20 of the patent, wherein the providing step additionally includes providing a color image of the region of interest to the image processor. -97- ^ The paper size is applicable to the Chinese National Su Yc called μ specification (2 Euro π 7 male directors) ~ 4 # 4104 ABCD VI. Applying for patent application 25. If the method of applying for the scope of patent application No. 20, the automatic processing step includes applying at least one image processing method to a user-defined area of interest. The method and application to The image processing method of the area of interest that is automatically defined by optically viewing the object is different. 26. The method of claim 22, wherein the providing step includes identifying a defect type in the region of interest, and the automatic processing step includes applying an image processing method suitable for the type of defect in the region of interest. 27. The method of claim 23, wherein the providing step includes identifying a type of type feature in the region of interest, and the automatic processing step includes applying a type of type feature type suitable for the region of interest. Image processing method. ~ μ 28. If the method of the scope of application for patent No. 20, wherein the at least two interest areas defined by the user 'are optically viewed at the software-defined step, then the object is automatically viewed by the optical inspection of the object automatically The defined prestige area of interest is automatically defined in the hardware inspection step by optically viewing the item, and the automatic processing of each image of the 2 area of interest is performed in the software image processing step. 29. A method for inspecting a spherical grid array substrate, the method comprising: at least a reference image of a product train substrate; a pattern of signing; and storing the pattern in a memory; obtaining an image of the spherical grid array substrate ; And whether the image pre-region feature of the ball grid array substrate is suitable for this mode. ^ Determined-Please read the precautions before proceeding. B. Page printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs. -98- 外 U outside A8 B8 C8 D8. Application scope 30. If the scope of patent application item 29 Method, wherein the feature includes a circle. 31. The method of claim 30, wherein the pattern of the circle includes a radius within a predetermined tolerance of a center point of a predetermined position. 32. The method according to item 3i of the scope of patent application, wherein the parameters of the mode can be adjusted at least partially in offline mode before inspection. 33. The method of claim 29, wherein the feature includes a solder pad. ! .IIITI · !! Packing · ί (Please read the precautions on the back and then on this page) Order · • Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs -99-This paper size applies to China National Standard (CNS) A4 (210 X 297 mm)