TWI303392B - A method and system for automatically focusing an image detector for defect detection and analysis - Google Patents

Description

1303392 IX. Description of the invention: [Technical field to which the invention pertains] The present invention relates to a machine inspection, which is a production and execution of a focus interpretation method and system. One μ Auto [Prior Art] There are a variety of imaging systems (4) _g systems that can be used to focus on them and maintain their focus as the object moves relative to the image system. The autofocus method in the present invention is used for liquid crystal and flat panel display, printed circuit board, L MEMS domain line, semiconductor component/wafer, biomedical sample, etc. for testing/viewing and adjusting image _ Fine detectr, such as wire ^ spacing to customize the focus. The following discussion will describe an autofocus system in a system that uses an image to move the image relative to the target object to adjust the focus plane, but can also be applied to the igurations. As detailed as the coffee and said that he (10) in the 2nd and 4th of January 2nd, 4th of the year, entitled "Methods and Systems Employmg infrared e-gmphy for Defect Detection and Analysis," The US special 10/3 is called _余, which describes such a system, which is referenced in 4 columns. I]. The autofocus method of the machine inspection system can be broadly divided into two main types, Position sensing _iti〇n sensing) and content analysis (c〇ntem analysis). The position sensing side is to measure the image gamma jf| and the distance of the object, and focus according to the stunner. j typical inspection system The position-sensing autofocus method rotates a constant distance between the object and the image σ. The position-sensing autofocus system usually includes a feedback control system 1303392 that includes σ, private animal parts and/or imaging systems to maintain the correct distance. The TeleTracTM Laser Tracking Aut〇F〇cus from Santa Barbara (cA) is an example of position-sensing autofocus. ......... The position sensing method has pure disadvantages. .first The distance maintained by the autofocus control (4) must be adjusted to coincide with the best focal length position of the image detector. Bean a, /, 夂, target object only - a small part is used to make the distance, resulting in the object The rest of the bets can be well focused. Third, the distance _ amount may not be suitable for some objects; for example, if there is a hole in the piece, the laser-based distance detection may not be applicable. The analysis method can achieve and maintain the best difficult position by capturing the image acquired by the image system. Each of the series of images using the scale method applies a focal length measurement function = f_-m_nng (4) (4). The resulting focal length measurement set is analyzed to determine the optimal pair. The secret content analysis method is based on the image and some minor indices of the pure point distance, so the content reading method can overcome many of the methods The shortcoming 'and the best focus position can be calculated. In the typical money-checking towel, FMF can be based in part on the distance between the image side device and the object. Difficult to work as a function of distance, you can get a – weaving measurement curve, where the peaks can be correlated with the position to find the best focus position. The age of FMF should have the following main properties: 1. Single mode within the relevant range (unimo catch (7), or single peak, maximum or minimum value; 2. accuracy of peak relative to the best focus position , or coincidence; 3· reproducibility (reprodudW%), or a sharp peak; circumference (for example, FMF# in a range of focal lengths in which the direction of the best focus position is determined without blurring); Applicability, or the ability to act on different types of images; 1303392 6. For the parameters that do not impact the focus, the insensitivity of the changes, such as the variation of the average image intensity; Or use image analysis to analyze the same image signal; and 8· speed: it should be able to calculate FMF at a faster rate. "For additional background information on FMF, see "ch〇i and focus technology" ( "Focusing Technique,,,J〇umal 〇f 〇pti heart 2824-2836, Ν〇ν·1993), and US Patent No. 5,932,872 to Price, both of which are referenced in this ship. The system is summarized as follows , Focused-image, at point (4) with / (χ, minus, is defined as a single exposure H & with respect to the direction of (^) from the object point, and the person shoots to the entrance 瞳 ( Entrancepupi is the total light energy of the aperture of the camera. When the image fails to focus, the weight of the single point on ^η〇ύ〇η (or the degree of dispersion W is a real model of psF: ,, i^2+72<i? 〇otherwise (1) where the system is a fuzzy circle> ψ PSF, where U is + diameter, and the ideal focal length corresponds to (10), ... know, less) The delta function. The fuzzy radius is as follows:

R

(2) - The middle D is light, and the focal length 1 is the distance between the lens plane and the imaging surface of the image detecting crying 1393392 (for example, the focusing plane array), and w is the lens plane and the object flat mosquito. distance. In this style, it is positive (if the silk image is recorded after the female image) or negative (if the silk surface is attached to the silk image). In the actual image of the towel, to use the mobile shadow to lose.) ___ both of them - can be secret. For an ideal focus image, this gives the result. The Fourier transform (F〇uriertransform) of PSF Λ〇,3;) is called an "Optical Transfer Functi〇n". In terms of the larger radius of the fuzzy circle, 〇TF will Attenuation of more severe high spatial frequencies (sharpness) in the image/(in other words, a larger __ indicates a more blurred image. The sequence can measure the sharpness according to the high spatial frequency content of an image and Focusing on it. In addition, you can use the high-pass wave (linear or non-linear) at the image of each image, and take the average intensity, value (or average energy) of the image after the image, and get an indication. In the high-altitude _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ The measured value can then be analyzed, and the best position is difficult. Several pieces of ginseng robbed how to get the confession (four) system, the target object 'and the degree of magnification to identify the best focus Method of measuring the focal length of the position. This References comprising: a first U.S. 5,231,443; 5,932,872 *, 6, 〇37,892; 6,201,619 '6,463,214; 6,151,415; 6,373,481; and Patent No. 6,222,588; and published Subbaro and Tyan ieee Transactions on Pattern Analysis and

Machine Intelligence, Vol. 20, Να 8, pp. 864-870, Aug· 1998, entitled “Selecting the best focus measurement for autofocus and focal depth” (“Selecting the 〇ptimal”

Articles of Focus Measure for Autofocusing and Depth-from-Focus": These documents are hereby incorporated by reference. A conventional system for applying computer-based autofocus methods is from TNP Instruments Inc. (Cars〇 n, CA) The Pr〇beMagicTM system, and the σ 亥 system use FMF's so-called Roberts traverse gradient cross gra (jient) !3〇3392 to perform autofocus. The method is to move the focus motor up and down to Determining the best focal length. One of the disadvantages of this approach is that it uses a single filter q (Roboz) that cannot be adapted to multiple images. In addition, this approach does not include adaptive autofocus, ie The autofocus is not real-time. There are many currently available methods that are not suitable for different types of images. · This is a serious image system for viewing unspecified images. Defects. In addition, adaptive m〇ti〇n-contr〇l algorithms are usually implemented in the form of electronic or embedded microprocessors. Therefore, it is difficult to reconfigure images of different types. Therefore, a fast and accurate autofocus method is needed, which can be easily adapted to different image systems and different types of image applications. Cold autofocus method can be applied to Used in automated systems for defect detection and analysis, such as 'electronic circuit such as printed circuit boards (PCB), integrated circuits, and flat-panel displays (FPD), which can utilize infrared (five) heat maps Video is tested to find defects. Typically, the power supply is supplied to the device under test (DUT) to structure its various components (device plus heat. An infrared detector can then capture The test image of the heated DUT. The resulting image, ie a set of pixel_intensity values spatially associated with the acquired image object, can be compared to a similar set of reference image data. Reference _ poor material _ difference, usually referred to as "a synthetic image" ("eompositeimage" person can refer to Defects are identified. Defect recognition interpretation can analyze synthetic images to automatically identify defects and thus increase the push rate and quality of 70 pieces. Examples of such inspection systems include, but are not limited to, , FPD, PCB, MEMS-based components, semi-components, and raw samples. One purpose of such a system is to detect defects that may occur on the component at certain points during the fabrication of the test component, and if it is identified, the defect can be repaired by the repair system, or The component can be culled, with 10 1303392 achieving manufacturing cost savings in both cases. Other applications include research samples, such as biological samples, in which the inspection and identification of artificial structures. A particularly important application of IR thermography is the active layer of a liquid crystal display (LCD) panel, or the "active plate". Using defect analysis can increase process and increase manufacturing yield. Equally important, as long as the number and extent of defects are not too large, defective panels can be repaired, increasing manufacturing yield. Figure 1 (Prior Art) shows a portion of an active panel that can be applied to an LCD panel. (Figure 1 is taken from U.S. Patent No. 6,1,424, issued to B.Sacchi on August 29, 2000, which is incorporated herein by reference). The active board includes a first shorting bar connected to each pixel HO of the pixel array via a set of source lines 11$, and is connected to each pixel via a set of questioning lines (control lines) 125. A second shorting bar 12 of no. According to Bosacchi, the active board 1 is used to evaluate the IR radiation of the active board excitation in the case where the voltage is supplied to the shorting bars 1〇5 and . In the case where the power source is supplied as such, the local portion of the plate excitation operates as a resistive circuit, and heat I is dissipated. Next, the heating reaction characteristics of the plate excitation are evaluated, and it is preferable to achieve a stable operating temperature at the plate (10) (heat balance is discussed. In the absence of the defect, the entire pixel array should be heated evenly. The bamboo is abnormal. The uneven scale sign of IR intensity is rewarded, _ can indicate the existence of gamma. The reference intensity value can be obtained by averaging the average image value of the image frame, or the ideal or defect-free reference plate. A reference frame shows that -f knows the part of the pixel (10);;, the defect in which the number can be =. The structure of the pixel 110 shown in the figure is related to the main function of a liquid crystal display. Including the finishing scale 115 m - the processing end, ^ the gate line - the - control end 'to the thin film transistor 20 料 to the _ capacitor 2 years of processing end. Le one of the defects, in the picture In order to explain the nature of the package, 11 1303392 includes both short-circuit and open-circuit. The short circuit between the two appears in the source line us and the interpolar line: 125 (short circuit 215) or common line 212. (short circuit 216); two current processing terminals of transistor 200 (short road 2 2〇); the gate of the transistor 200 and the second current processing terminal (short circuit 225); and the two ends of the capacitor 21〇 (short circuit 226). The open circuit splits the source, the gate, and the common line into segments (open circuit) Milk, 228, and 229), and the open circuit between them appears in the source line 丨丨5 and the transistor 2〇〇 (open-way 230), the gate line 125 and the control end of the transistor 200 ( Open circuit 232), capacitor 21〇 and common 'pass line 212 (P exhaust 235), and between transistor 2〇〇 and capacitor 21〇 (open circuit 233). Each defect in Figure 2, plus other numbers A defect that adversely affects the operation of the pixel 110. Unfortunately, many of these defects are difficult to detect using conventional testing methods. Therefore, the art towel requires an improved method and system. Identification and localization of deficiencies. Some inspection systems include an excitation source that energizes the object under test by means of a defect-to-image system. The type of excitation is The image is secret, where the imaging system may be based on visible light, infrared light, combined spectrum, magnetic field, etc. Obtaining the image. Regardless of the image system used, the two knee images of the object being tested are compared with a reference image to obtain a composite image: significant difference between the test and the reference image. It will appear in the synthetic image to identify possible defects. ^ Some forms of excitation will produce artificial artifacts of defects, which are caused by defects between the test and the reference image, but not true f and The defect areas are related to each other. The short circuit between the two lines increases the flow through the record, thus increasing the temperature of the lines along the short circuit. Thus, the lines themselves are not defective. However, it will appear in the composite image along with the short circuit. It represents the defect data of this short circuit, so it is hidden in defect_artifact data (ie, a "defective artificial sign" ("secret (10) (4),)). Defective artificial signs often obscure the relevant The flaws make it difficult for 12 1303392 to be accurately positioned. Human operators can use the careful study under the microscope to remove the mosquito in a defective artificial sign, but the speed of the person is usually relatively slow, and the report is exhausted. Therefore, there is a need in the art for a method and apparatus for automatically distinguishing defects from artificial fingerprints associated with them. [Summary of the Invention] - This is an autofocus interpretation method and system for machine inspection applications. The production/image format of the implementation/Guangze method (visual, infrared, spectral, scanning = shadow = _II type has no M. The image technology disclosed in the present invention, the image j giant measurement function and the adaptive speed The control method can be applied to the computer-based inspection system, which can be used for the application of the degree of amplification. Seized the money system of the existing imaging hardware, ^ and does not require any additional components.

In some embodiments using infrared thermal images, the test vector _vector will be examined, and the structural features on the piece are heated to produce useful thermal features when identifying defects. Test: The I tree is applied to enhance the thermal contrast between the features of the defect face, and the money IR image is also known to obtain the heat map image from the X. The mathematical transformation applied to the application of the aforementioned autofocus transfer method can provide an improvement in the analysis of the lack of shame. Material defects: f occlusion of the image of the person who is missing the image, or "defective person money", causing difficulties in the processing of defects. Some implementations use defect location interpretation that analyzes the defector's money image to accurately locate the corresponding defect. [Embodiment] FIG. 3 shows an inspection system 300 in accordance with an embodiment of the present invention. In this example, the system includes one or more suitable image detectors (infrared camera fader - video camera 3K) that point to the target object 315 for image acquisition. A shadow 1303392 processing and control system 32G captures the image from the fields 305 and 310 via the frame of the 知(4)(4) milk. The image_device 3〇5 and the training include the two-dimensional domain structure, or the surface structure is supported. In this example, the second self-control action control || gamma directional and the motor-turned Z stages 335 and 340 send control signals to change the distance between each camera and the object 315. A ♦, mechanical motor driven χ γ γ 台 台 柳, which is also a computer controller, the object is moved within the line of sight of the camera. In the embodiment, the system 32G is a general-purpose electronic money in the form of a software, represented by an image analysis module milk and a motion group 360. / city m. In response to an instruction from the motion control module 36, the motion controller communicates a trigger signal to the frame grabber 325 to coordinate X, γ, and 2 loads with the captured image. The module 355 analyzes the captured image and, if the image is deemed to be a deviation from the stool, issues an instruction to the motion control module (10) to change the speed and/or direction of the appropriate stage. & According to the embodiment of the present invention, the image analysis module and the group 355 calculate F· independently of the type of the camera, or the type of the image. This trait makes the embodiments of the present invention extremely important in the inspection of the machine inspection system, in particular, the cable and the video 'available for measuring the tablet ship, the pcB board, and the job S is the basis. Line 'casting components / wafers, biomedical samples, etc. ‘、、、土 FMF Selected ^ In some cases, the image analysis module 355 incorporates a flexible universal use-group* pass (sharpness) filter. For the specific high-pass filter, the FMF architecture of the filter is given by: () ^(ι) = |fc][ σ (3) where I is the acquired image, #(1) is the image The average pixel intensity, while 丨14 1303392 and σ are parameters that allow for flexibility when looking for the best FMF. α is usually 1 or 2, and is the power of the filtered image (that is, the filtered image may be squared) 'CT is a scale factor (usually 1, 〇〇〇) . Although not necessary, and either 7 or both can be a function of the amount to the left. In this case, the FMF architecture ~(1) can be further optimized by using the change function. The main difference between the different FMFs of this architecture is the 滤波-filtered state φ(). The flexibility and adaptability of this architecture is an advantage of some embodiments of the present invention, as will be apparent from the following. The following paragraphs of the description text will define and list the filtering used in some embodiments. A detailed discussion of it and its additional filters can be found in textbooks for image processing and in technical manuals showing the art (for example, National Instruments, Austin, IMAQ Vision Concept Manual, which is hereby incorporated by reference. ). Figures 4A and 4B are a pair of sample images 4A and 4〇5 which are used to illustrate the method by which certain embodiments of the present invention are based. Image 400 shows a periodically magnified periodic structure of the substrate under test, while image 4〇5 shows a fiducial mark on the same substrate. Figures 5A, 6A, 7A, 8A and 9A are plots of the focal length measurement as a function of Z position for image 400, respectively. Each of the figures uses a different arrangement of the 32FMF structure. Figures 5B, 6B, and 7B, and the offense are plotted against the image 405 of Figure 4B as a function of the z position for the image 405 of Figure 4B, respectively. Corresponding to each _ group map (for example, 5A and 5B), the same arrangement as that of Equation 3 is applied. The various arrangements of Equation 3 employ different filter φ as described below, and a reference value of σ approximately 丨^(8). The numerical values of the parameters are shown in Figure 5Α/Β, 6Α/Β, 7Α/Β, and 8Α/Β in the FMF, and Figure 9Α/Β in the FMF. These figures represent the sharpness of the image as the focus surface (for example, the distance of the camera); ideally, the peak mine should be the best. The linear Laplacian filter is the most common linear test used for autofocus. In the discrete media (di_e vice (five) towel, we pull g pull " kg / wave is defined as a whirl of tears with the following core (kemei) 15 1303392 convolution filter: Η M kll Η W · (4) Η Μ H (4) where x = —2^l + W+kl + M). In particular, the simplest finite difference analogy of the continuous Laplace operator is the following core: 0 1 0 1 -4 1 0 1 0 (3) As shown in Figure 5A, the FMF of the Equation 3 is imaged. 400 is not a single mode. A built-in autofocus system to which this filter is applied may select a Z position corresponding to the peak of the error and thus may not focus properly. This deficiency is a balance of using a non-linear filter, although it has other advantages, for example, it usually requires less computation. A non-linear Roberts filter can depict the contour to indicate pixels that vary in intensity along the diagonal axis. The filtered value of a pixel becomes the maximum absolute value between the change of its upper left adjacent region and the variation of its other two other regions, which is given by: (6) where the imitation refers to the image I (i, j) the intensity of pixels. This filter is widely used in the automatic (4) system of commercial machine money checking products (for example, (4) (four) face coffee,

Ca=0n, CA's Pr〇beMagic). As depicted in Fig. 6A, such a wave device, which is an early nuclear state, has an unfavorable "flat" zone. The chopper in Equation 6 is therefore not the best choice for image 4〇〇. By the way, the nonlinear difference filter produces a continuous contour by using each of the intensity variations between itself and its three upper left regions. The value of a pixel filtered by 16 (7) 1303392 becomes the absolute value of the maximum variation in the upper left area. , ./-1 As shown in Figure 7A, this filter produces an unfavorable flat area 7GG with area 6 of Figure 6A, which is therefore not the best choice for imaging. , the nonlinear gradient filter, the wave material will be repaired along the (four) straight axis where the intensity changes. - The new value of the pixel becomes the maximum absolute value between the change of the upper region and the two left regions: (8) The chopper applied to the image_ as depicted in Fig. Single mode and no flat areas in Figures 6Α and 7Α. The filter of Equation 8 is better for the image 4 〇〇 ^ application than for Equations 6 and 7. Figure 9 shows a tf image 400 derived from the filter of Equation 8 and similar to the graph of Figure 8α - FMF graph. However, the graphs of Figures 9Α and 8Α are still different, that is, the graph of Fig. 9Α (4) (remember the graph of Fig. 8Α (4)). Comparing the graphs in Figures 8Α and 9Α, the former has better reproducibility, ie, the sharper maximum value, while the latter has a broader outline. Depending on the autofocus method used, the reproducibility is more important than the width. As detailed below, the benefits of scanning autofocus methods for reproducibility tend to be more than the benefits of width, while gradient-based interpretations tend to achieve greater benefits than reproducibility. Sexual benefit ^. Part of the benefit of the FMF architecture in Equation 3 is that it is adaptable to the suitability of a given system. Other commonly used filters include nonlinear PlyWitt (N〇niine read e she), non-linear (four) gamma (10) nllnear Sigma) 'and nonlinear Solang (secret wheel s. (4), etc. as described in the present invention Among the choppers discussed, the nonlinear Loboz and the nonlinear gradient 17 1303392 are relatively simple to calculate. The Roberts is more symmetrical than the gradient, and it is often a better choice. However, since the nonlinear Loebz has a "flat" area at the off-focus position, the nonlinear gradient filter is therefore preferred for the image 400. The discussion of Figures 5A through 9A illustrates The flexibility and adaptability of the disclosed FMF architecture allows the autofocus method disclosed in the present invention to be adapted to a variety of images and systems. In addition, the same autofocus software can be parameterized and Applicable to a variety of image acquisition types (eg, infrared and video), magnification, image type (as shown in Figures 4A and 4B), and many other important properties in machine inspection systems. The description describes in detail the method of obtaining an acceptable focal length measurement; the following will describe in detail how to use the focal length measurement to achieve and maintain optimal focus. Using the FMF architecture proposed by Equation 3, the present invention reveals this. Two techniques. Although the two techniques are better, 'but the foot is not limited to this. The first technique' scans the best focus position & is nearby, scanning the camera and taking pictures. The transition between objects includes the check of the button. In the scan _, the value of _ = is compared with the maximum focal length measurement, or the value is 1 or 2 techniques, based on the gradient, the lake, Adaptive motion control to find and turn the peak focal length measurement value., Come and give = kind of practice ί 'the focus position is carefully measured for the fixed object 疋' and in the two brothers, the object In fact, the focus is maintained, even when there is a movement in the object. The first method has its own movement in the test program, and it must be in the hand, and in the _ law is in the hand t or other machine 11 It will be implemented before the visual system. The second method can be Pour = and turn around. In this case, the interpretation should be read, and the operator or control (4) can be 18 (S) 1303392, change hardware, etc. 10 is not suitable for automatically selecting one of the best FMF for a given image. The dynamic focus translation method 丨 _. First, in response to the instruction from the module, the action control is 330 to the Z stage. 335 moves to a starting position (step 1〇〇5), and captures an image of the target object (step 1G1()). The motion controller 33q then repeatedly pushes the z-table tear (step 1015) and captures The image (step _ until the Z stage 335 reaches the terminal of a predetermined scanning range in which the A is the best focus position of the false e again (decision step 102). } Each captured image can be forwarded to the module 335 for analysis by the frame grabber milk. The TMF 'such as described in Equation 3 is applied to the image collected by the step 1010 and the name is collected (step 1〇25), and the result is the state as depicted in Figure 8a. - A focal length measurement curve ugly. The image analysis module 355 analyzes whether the curve is wide 1030 (step Cong). The face can be provided with a suitable (four) basis for autofocusing on the object to be imaged. In one embodiment, the step Cong records the number of peaks - the sharpness of a single or multiple peaks, the width of the graph _, and the degree of noise. According to the decision step, if the graph 1〇3〇 cannot meet certain sub-standards (for example, as shown in Fig. 5A, there are two peaks), the other-FMF will be used for the image acquisition program, and the step ship 5 and _ are repeated. Different mFs are tried (step purchase) until the decision step _ determines that a certain graph result can meet the minimum time of autofocus. In other implementations, several FMFs are applied to the image taking material and the curve that best matches the target image is selected. After selecting the recording focal length system (fourth) and meaning, in step _ and group 355, the position of the z stage corresponding to the maximum focal length measurement value is recognized as the position surface. (In the higher noise image, the face-to-face approach may be to find 360t to find out the towel d is not the maximum focal length value.) The motion control module picks up the party's indication from the position of the module 355, and, as a reaction, 19 1303392 shows that the motion control is 330 to move the Z stage 335 to the focus position 1 (step 1 Still 5), leaving the image detector 305 focused on the object 315. For the sake of explanation, in Fig. 10, the acquisition of images and the analysis of images are separated. The actual system saves time by performing image analysis and image capture procedures in parallel. In some embodiments, the list of image types and their associated FMFs are maintained; in this case, the flow of Figure 1 can be selected in step 1025 for a known type of image - A preferred FMF, then jump directly to step 1〇45.

The main difficulty in the flow of Figure 10 is the action and image acquisition. However, its calculation is straightforward. The size of the _ which is taken is reasonably small (~2〇〇_300), and it is the most important to find the FMF dragon (that is, the most read-in system is effective. This method also Eliminates the need for single mode, convexity, and initial guess values in the function optimization program. Figure 11 shows that the 4-way Turan's system is based on gradient-based, adaptive, and true The method of autofocusing is the same as that described above with respect to the description of Fig. 3. The software component (module and 36〇) analyzes the local behavior of the focal length measurement' and determines Acquiring and maintaining the abundance of each of the stages of the langf.

-..., and 113〇), j follows - side training (blocks 114G, (10), said, (four) and U8G). The system will, for example, react to the wheel that stops the loop in response to the - fairy command or secret reset. Before starting, the control system 32G accepts the start position of the Z-stage, thereby the position? Begin to capture the "image", a reduction _ enchantment, (10) speed wire step (10) move Z stage 335, and a time increment τ 1105 separating the image capture step. This initial position can be the position selected using the translation plane of the graph. The program is initiated by step im, at which point the z stage 335 will have one more starting position for the image side device 3 (for example, located at the assumed maximum Above the good female position). Control system: 1303392 Controls the movement of Z stage 335, moves image side device 3〇5 to a starting position (step 1_ 'and obtains the first image (step paste. Image analysis module milk then facilitates f«) Like the silk-FMF ridge test, after the step 1130 or during the period, the action controller 33 causes the z-stage milk to start moving toward the best focus in the direction indicated by the data _. (Step 1 (10)) This action will continue for a time increment Δί, after which the frame grabber 325 captures the next image (step 1150). > In the simple steps, the image analysis module 355 utilizes the nearest The captured image % is again measured by FMF using a FMF. The module 355 is then calculated by two or more new focal lengths (step (10)), the eye level, and the time increment mosquitoes. The focal length system is a measure of the sum of the gradients. The image analysis module 355 then calculates a new focus plane adjustment speed and/or time increment based on the gradient (step 118〇). The following describes the calculation of several new speeds and time increments according to step 1180. Method control = translation method. Participation 3, interpretation of an ideal hypothesis ° 335 speed). In this case, the dynamics of the Z stage 335 is as follows: dz (9) dt where v is the step 1180 The ^γT-connection degree of the library m to be calculated (figure U), and may be determined by the gradF (v) = shape of the focal length measurement F (by) /&;; (10) ί vv(gradF(z ?r)) The assumption of another - _ is π stationary, that is, it does not significantly depend on time t and ^ in this case, the stage should move in the direction of increasing FMF, that is, v The payout should be the same as the gradient ship, and the value of the v in the material case is proportional to the gradient of 1303392: v = /?gradF(z) (1)) where /Η is a coefficient of proportionality According to several embodiments of the present invention, this robust simplified model can provide the basis for the development of several real-time autofocus interpretations. The effectiveness of such translations depends in part on the recent use of computers. Progress on the basic motion controller (for example, the brain (five), he,

Aerotech, etc.). These controllers can provide an excellent service loop and can be scaled with very reasonable precision, such as those specified in the aforementioned simplified surface of the present invention. As long as the image _ can always remain within the range of focus, that is, within a single attraction area of the full-scale peak (non-stationary) of the focal length measurement, the time variation of the 量 distance measurement value ( It can also be ignored to some extent because of the change in image. In the controlled interpretation method that is not disclosed here, the speed is selected as the fragment value ece: constant). Control system 32 can change speed V at any time. Thus, the interval of a g疋 velocity is as follows: _ outside _ outer color = color % m, and dt v ν ^ /, ' is the time for acquiring the image and then calculating the second focal length measurement value ^. This method is a repetitive person, so it only describes the process 1 map) 115〇), Ye Zeng ^ in the mother's repeated, the system 300 obtains an image (step i ^ different i The focal length measurement value (step 1160) and the focal length measurement value gradient (step-slip update speed v, and/or time increment Δ, (step). At the time of the second and second restart, from the previous The velocity Vi and the focal length measurement value of the helium pole will be recorded. In U, the control interpretation according to some embodiments of the present invention is first described, and the description needs to be used.

22 1303392 Some symbols. The present invention defines a function ·· range Ιά^Λ->ά(13) and the following quantities·· ζ, Δζ> _ stage in the stepper motor count [c] · privately - the stage is at the number of revolutions [ Relative position in r];,, relative position, 仏|〇卜级/,] handsome], where service /r] is the stage t, At, tt, 屯 - time in seconds [s] and Time between ^ huge; number of tips; ν, ν〆' - in revolutions per minute [r / call or 皙 vcps - the speed in the count per second [c / s] Edit her to replace the FMF unit _ the focal length measurement value of each S method (four) calculate the 撷 及 及 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ (匪) Figure 12A to 12D are four gradient-based autofocus interpretations of the same image, the focal length measurement value FM and the velocity v versus the time in seconds. . Each of the translations uses the same FMF as discussed above in connection with Figure 8A. The upper graph in each figure shows the focal length measurement when the motion controller 33 brings the image detector 305 into the in-focus position, where the initial detector position corresponds to the zero position on the inter-scale. . As the flow of Figure u progresses, the focal length measurement increases as the image is detected as approaching the best focus position. The graphical representations of Figures 12A through 12D show the relative performance between the various control translations as will be described in more detail below. Autofocus Interpretation #1 (Basic Gradient, Fig. 12A) 23 1303392 Given (input) degree calculated distance (the movement within this distance from the focus position should not be this) > like quality can be noticed Change...

Axv0[RPM] - Initial speed. Magic a\ r c μ rs The coefficient of speed calculation. Conversion factor from [c/s] to [r/m] Zero repeat (start): 2. 1· $speed V. Start (or change) the action. Get the image and calculate the focus measurement. / Repetition: 1. Calculate the private μΑχ and wait for seconds. 2. Acquire the image and calculate the focus measurement value. Ten ^ Γ, and the stage speed is changed to this value. 4. Go to /+1 repeat. Ζ This is a straightforward gradient interpretation. For the current speed ~ coverage distance & required ^, the potential drawback of the translation method is that when this method is accepted for a long time, the autofocus interpretation method #2 (qualified ^, figure 12B) Given (input): Strong in this turn __ should vJRPM] - initial speed. 》 rc μ r s cm The coefficient of speed calculation. Conversion factor from [C/S] to [r/m] 24 1303392 △, max [s] and [s] - Δί limit values. Zero repeat (start): 1· at speed V. Start (or change) the action. 2. Obtain the image and calculate the focal length measurement value / 7. . The first/time repeat: times ( Αχ λ 1. Calculate = range|^min, j^, △, and wait for the second. 2) Obtain the image and calculate the focal length measurement g. 3· Counting differently Change the stage speed to this value. 4. Go to /+1 Repeat. Here 'is limited to [&min,&maJ, and the difference is -丨 is divided by (V! to calculate the gradient) The center is the speed calculated in counts per second. One potential disadvantage of this thickness ten is that 'the speed value is not limited, so the minimum speed can be used to calculate the gradient of the noise. If V/& max«〇, and very large speeds, the stage may be moved too far (if v^min »1). Except for these areas (ie, when it is neither too small nor too Great time), this translation can show its best performance. 'Autofocus interpretation #3 (limited | v), Figure 12C) Given (input): Heart W - Gradient calculated distance (by focus) The movement of the position within this distance will cause noticeable changes in image quality). ...

V〇[RPM] - initial speed. a——-The coefficient of speed calculation. Mu_ r s A - *[c/s]S[r/m] conversion factor. U 1X1 Count 0 Zero Repeat (Start): 1. At speed V. Start (or change) the action. 2. Acquire the image and calculate the focal length measurement value F. . No./Repeat·· 25 (S.) l3〇3392 1. Calculate △[·=# and wait for △[·seconds. K-il 2. Obtain the image and calculate the focal length measurement f. 3. Change the stage speed to the value calculated according to the following steps ^ : a· Calculation 〇

AxsgnvM b. Calculate hhrange^Ln-like |, |v|mJ 〇 C· Compute '=Bu,. §11. D_ Go to /+1 repeat. The main feature of this interpretation is its limit on the absolute value of speed, but it also maintains a positive sign. In this case, △ "within the limits of awning, fragrant. 4 X 丨丨min experiment shows that this translation is in a reasonable range of performance. (B). However, a potential problem is In the whole range of the plant value, the value of the good can be = ^ the parameters of good work exist. This translation method may be most useful for specific sub-images. The amount of focal length of the image is not the same as the _ parameter, such as 4B The one shown in the circle. ^ ^ The interpretation method is based on the gradient symbol ' instead of the gradient itself, so it does not depend on the range of the focal length measurement value. ' Autofocus interpretation method #4 (常定|v" = r , Figure 12d) Given: Heart [c] - Gradient calculation distance (no change in image quality caused by the reading of the dance position). , , , and set the movement within this distance should be F [RPM] - the absolute value of the speed. The coefficient of ab^_lβ velocity calculation. Θ 'Conversion factor from [(^] to [1*/1^]. Zero repeat (start)·· 1. Calculate A. v 26 (S ) 1303392 2. Start with speed + ( (or change) Action 3_Get the image and calculate the focal length measurement value F. The first repetition: 1 · Special treatment Δί seconds. 2·Acquire the image and calculate the focal length measurement value 6. 3. Branch 1 shout (five) and the stage speed Change this value (if 4·into/+1 repeats. A potential advantage of this translation is that it does not depend on the range of FMF values, and it is imminent in the gradient - a minimum (Good behavior at the point where the attraction area of the _ is not in focus. The cost is that a higher speed can be tolerated for a relatively large gradient in the same focus range _ _ the same speed (7). The aforementioned interpretation is shown to maintain focus during movement of the object in the χγ plane. System 300 (Fig. 3) captures an image 13〇〇 for an area of object 315. ^ Stage 350 is followed by five (four) coffee Moving at a speed, this is the speed at which some types of inspections are seen. Before moving, the autofocus is turned off and the object is still tilted relative to the surface of the load 350. When the object 315 moves horizontally, the image is milked out of focus. After the money 300 is followed by the original image measurement, the tester is repeated, and the scale (4) is opened. The result image and image 1305 _ contrast can be the ship's hair - revealing the method of having a female. Figure 14 shows the dynamic focus applied to the AF in Figure 13, when the XY stage 35 〇 moving the object 3b between the coordinates of the dynamic reaction. Close The lower focal length measurement of the destination (<2〇) is the result of the sudden stop of the stage, rather than the instability of the interpretation. The application of the defect detector f ^ shows a test system , _ _, which includes the well-known board please 5 and according to the application of this month, the system is inspected. The board (10) 5 series and the drawings and 2 of the board just like 'eight in the same or similar numbered components Mark the same or similar components. Board 15〇5

27 S 1303392 includes a shorting bar 1512 not shown in Figure 1, but it is still conventional. The inspection system 1510 includes an IR detector 1515 (e.g., a Han camera) that is positioned above the board 1505 to provide image data to the computer 152 via the frame grabber 1525. An excitation source, The signal generator 1530 provides an electrical test signal 'or a test vector' to the board 15〇5. The test vector adds heat to the structural features of the board 15〇5 to generate a defect for identification. The thermal characteristics of the computer 1520 control signal generator 153 施加 apply a test vector to the board 15 〇 5. These test vectors can enhance the thermal contrast between the defect and its surrounding features, thus allowing the melon detector 1515 to obtain more enhanced The heat map image is used for the detection and analysis of the defect. The computer 1520 will further determine when the image data should be acquired, receive and process the captured test image data from the frame grabber 1525, and provide a User interface (not shown) IR detector 1515 should have excellent temperature sensitivity. In one embodiment, detector 1515 is a 256 χ 32 〇 component (recorded indium,

Antmomde) an IR focus plane array thermal image camera (ir - secret coffee)

Array Thermal lmaging Camera). This photography _ minimum temperature recording is below ο·degree c. In some embodiments, a plurality of IR detectors are included, such as relatively low magnification cameras for defect detection, and high-amplitude IR cameras for defect side and analysis. Additional cameras can also be used to increase the inspection area, thus increasing the inspection bandwidth. The chemical generator 1530 supplies a source test vector & to the shorting bar 105, a gate test vector & to the short bar 120, and a common test vector Vtc to the shorting bar (5) 2. Referring again to Figure 2, testing of certain types of defects (e.g., open circuits 230, 233, and 235) requires transistor 2 to be turned on to create a signal path between the corresponding source and common lines (1) and 212. The signal 153 can be used to (10) supply the DC test to the s vTC to the gate line 125 by the shorting cup 120), so that the transistor 28 1303392 200 can be turned on, and the test vector is supplied to the source and common lines. And ^. Passing through the capacitor 21G will transfer the current. If the short circuit 226 does not exist, even if the transistor is biased in the forward direction, the undefective pixel (10) will not have a direct current through the U. The source and common line test vectors ~ and & are thus selected to produce an AC signal that can pass over the valley 210. The frequency of the Ac signal matches the load provided by the board (10). The bribe is on the 15th, and this accelerates the heating and thus speeds up the test. Placing the power transmission to the maximum also allows for a lower supply surface, so that less sensitive components are not damaged. It is also important that the details of the 'speed of contact' and the Mosquito time control set as detailed below provide an enhanced thermal contrast. In one embodiment, the source test vector ~ is from zero to 30 volts at a frequency of about 70 kHz, and the common line test vector & is at ground potential. Some embodiments of the present invention provide an AC or DC test vector between source line 115 and common line 212, source line (1) and gate line 125, and gate line 125 and common line 212. Other embodiments use an AC signal to turn on the transistor. The AC and DC test vectors are supplied at the same time as the above, and more complex tests can be achieved than the only waveforms (for example, 'only DC' AC, or pulsed DC test vectors). Figure 16 shows a portion of a board 1600 that provides enhanced testability in accordance with an embodiment of the present invention. The board 1600 typically includes an array of pixels 16A5 each connected to a source line 1610, a gate line 1615, and a common line 162A. Four sets of shorting bars (source bars 1625, gate bars 1635 and common bars 163 〇) allow an inspection system, such as that shown in Figure 15, to test a subset of pixels 16〇5. Alternatively, four sets of the same type of sad rod (e.g., four source rods or four gate rods) can be used to charge the selected column or row. When some features are recharged, the adjacent features are also removed, which can improve the contrast of the images. In other embodiments, only one set of one or 29 1303392 two types of shorting bars are provided. For example, it may only be that the bar 1635 is common to the _ bar: the bar 1630, in which case the pixels _ can be excited by being divided into four groups of columns. In addition: one: or more sets of shorting bars may also include more or less than four shorting bars. Figure 17 shows a portion of an LCD panel 17 in accordance with another embodiment of the present invention. The board 1700 typically includes an array of pixels 17A5, each connected to a source line 1710'-gate line 1715' and a common line 172'. The source rod 1725, the gate rod 173A and the common rod 1735 are segmented to enable the inspection system to supply power to the test area (e.g., 'area 1740'). If you follow a different approach, fewer types of sticks need to be segmented. For example, if the source and gate bars are segmented, the common bar can supply power to the zone 1740 without the need for segmentation. The area 174A may have the same field of view as the R detector used to capture the image. The number of pixels 17〇5 in a given area is typically much larger than that depicted in this simple example. Figure 18A is a graph 1800 showing a sample defect of a descriptive nature and its thermal response around the region. This sample defect is assumed to be a short circuit having a volume V of about 25,000 ohms of resistance R' defect and associated electrode integration of about 1 〇2 m3, and an exposed surface area A of about 1 〇-5 m2. The spedflc heat Cp of the electrode is assumed to be about 2.44 x 1 〇 6 J/m 3 K, and the convective heat transfer coefficient hair of the surrounding air is about 1 〇 w/m 2 K. For an applied power of about 6 microwatts, the equilibrium temperature at the defect is about 6.5 degrees C above the initial temperature. The following heat transfer model shows the thermal reaction of the sample defect in response to the power applied by Xin: VCp = PaPPlied ^ " T^ir ) (1 4) where: V is the integrated volume of the defect and associated electrode;

Cp is the average specific heat of the defect and associated electrode; T(1) is the temperature of the defect in seconds, according to the Kjelda count; 30 1303392 (1) is the power supply excitation applied in seconds; hair is the convective heat transfer coefficient of the surrounding air A is the integrated surface area of the defect and associated electrode; and Tair is the temperature or initial temperature of the surrounding air (eg, about 3 〇〇 κ). The formula (Η) actually means that the power supplied to the defective area is equal to the sum of the power of the god of the absence and the power of the environment that is diverging into the environment at a given moment. Initially, when the temperature difference between the defect area and the surrounding environment is the smallest, the first addition in the equation dominates the whole. As the temperature of the defect area increases, the second additive gradually becomes affected. " Figure 1805 shows the sample defect area 181A and its surrounding area 1815 as a set of squares, each of which represents the intensity of the image recorded by the image pixels. For the sake of convenience, Fig. 18A shows the defect area 181A as a single pixel. Figure 1800 assumes that about six microwatts is applied between the source and the gate for a time sufficient for the defect region 1810 to be a short between the source and the gate line - an initial thermal equilibrium temperature of about 300K. It rises to a final equilibrium temperature TEF of about 3〇6·5Κ. Under this particular condition, for a typical active board, the time to traverse the entire temperature window takes approximately —1 - 0.2 seconds. The strip reaction curve 1820 reveals the thermal reaction of the defect region 181〇. The vertical axis of Fig. 18A represents the temperature percentage at which the initial temperature of the reaction curve 1820 is between the temperature and the final equilibrium temperature tef. The horizontal axis represents time and is expressed as a thermal time constant of one. The thermal time constant τ is the time required for the temperature of the defect region 181 由 to rise from a given temperature to the final equilibrium temperature TEF by 63.221% of its entire range. For practical purposes, the defect temperature is set to the final equilibrium temperature after four or five time constants τ. The thermal reaction of the region 1815 surrounding the defect region 1810 will be different from the thermal reaction of the defect region 181〇. If the defect is a short circuit, heat from region 181 扩散 will diffuse into region 1815, 31 1303392 causing the temperature of region 1815 to rise with defect i8i. However, the temperature rise of the region 1815 lags behind the defect area 181 〇 and rises to a lower final heat balance temperature than the defect 181 。. The test vectors applied in accordance with certain embodiments may provide an increase in temperature contrast between the regions 181 and the coffee maker to allow the Chinese image to more easily serve the defective features. The IR inspection system then applies a unique image acquisition time control sequence to capture test shadows, like bedding, far before the defect area (8) reaches the final equilibrium temperature ΤΕρ. (If the defect area 1810 is an open circuit, the temperature of the area 181 落后 will lag behind the surrounding area 1815, but will gradually reach a final equilibrium temperature.) The graph 1830 of Figure 18B shows the defect and its surrounding area. The measurement and subtraction orientation and image acquisition time control between the enhanced heat contrast. The graph dislocation includes a thermal reaction curve Lu 184 ’ which shows that the defect 181G reacts to the test vector and is repeatedly heated and cooled. Figure MB further includes a pair of waveforms 1 "eight (^ and ΕχατΕ, which share a common day-and-nine scale with Figure 183. The high-side portion of the waveform iMAGE represents the region 181〇 and 1815 of the ir~ image during capture Time window. During each reference window, one or more reference images are captured, and one or more test images are captured during each test window. The high side portion 1855 of the waveform ΕχατΕ represents the test vector. It is known to add to the defect 181〇 to introduce the thermal contrast into the time period from the 181〇 to its surrounding area (8) 5. ^ In order to capture the test image, an inspection system (such as the inspection system 510 in Figure 15) is 1855 The test vector is applied to the device under test during the period. The inspection system then captures one or more IR images of the defect area well before the target feature reaches the final thermal equilibrium temperature. The day-to-day defensive may require the highest temperature (also, The peak value of the reaction curve 184〇 is maintained below 95% of the difference between the initial 1 degree and the end-of-end equilibrium temperature TEf. In some cases, the dish may even jeopardize the work of the DUt. The best upper limit of the overflow is therefore varied with different DUT 'test procedures, etc., but in many cases it should be better than 32 !3 〇 3392 86.5%. The experimental data suggests the highest temperature. Not exceeding 63.5% of the difference between the initial temperature and the final flat temperature TEF (that is, before a time constant is completed); that is, excellent results can be obtained. The heating/image taking steps are repeated. The results are averaged or integrated to reduce the effects of noise. The maximum and minimum peaks of response curve 1840 should be sufficiently separated so that the selected detector can resolve the temperature difference. ' Defect 1810 can be heated to The temperature in the south 'for example, lasts three or four times. The number of turns 'in this case' can still be taken back before the defect 181 〇 reaches the final equilibrium temperature TEp. Since heating takes time, choose a relatively low maximum The temperature still accelerates the test. In addition, it is selected to bring the image contrast to the maximum test voltage. If the application time is too long, the temperature of the areas (8) and (8) 5 may still be raised enough to damage. _ The degree of sensitivity reading. In this case, the application time of the test vector is sufficient to achieve the degree of thermal contrast but still does not raise the temperature of the area and the temperature of (8) 5 above some maximum temperature. In an embodiment in which the test vector portion is in accordance with the active plate of a liquid crystal display (LCD) having a resolution of the dish, the ^(10) wave side _ portion represents the time when the s vector is not applied. Secondly, part 1855 represents the time series of the AC source and the common test vector at a frequency of about 7 〇KHz from zero to 3 () volts _ s. The image intensity of a corresponding region. ^, the value is Also representative of the 1525 riding - butterfly, L money (10) 'frame grab strength value. ~ Like to send to the computer (10) - an array of pixels Some of the conventional inspection systems will have the effect of _. The images of the f-sequences are averaged to reduce the noise, and the image is then compared with a reference image to show the existence of non-defects. The foregoing method and system 33 (S) 1303392 of the present invention can be used to generate enhanced test and reference images. The present invention - an embodiment of the image 雠 in place of the conventional symmetry, will be applied to both the image and the sequence of reference images. The result is converted to image Ιτ and IR and then the difference is made. ^ test

Γ fr Hl > 1 1 LF ni ^ jj J (15) 1 D 1 represents an image conversion which converts the intensity of each pixel by the number of 16-bit 7L provided by the frame grabber 15 The reciprocal of the floating point reciprocal 1 (9 dishes 8^^!^186 1:0〇), below); 2· Γ is the first image of the sequence; 3· η is the number of images in the sequence; 4·F Represents image filtering, such as low-pass filtering, to reduce noise and eliminate data provided by impaired pixels in the 1515; 5. L is a search table for the image to convert the range of intensity values into one. Different sizes of fragrant seeds, for example, a different range or from linear to quadratic; and D is an image conversion (casting) from floating point values to sixteen bit values. When performing image conversion on a sequence of images (test or reference), the per-pixel intensity of each image sequence is converted to a floating point number. The resulting image array is then averaged on a pixel basis to integrate the images into a single image array. The resulting image array is then filtered to reduce the shadow of the noise (S) 34 1303392 and remove the data associated with the defective pixels in the imaging device. Each piece of data associated with a defective pixel is distinguished by an extreme intensity value, i.e., by a new interpolation intensity value from the data representing the neighborhood. The intensity values of the integrated filtered and post-image arrays are applied to convert the range of intensity values into a different scale, for example, a different range or linear to quadratic search table. Finally, the resulting values in the image array are converted from floating point values back to digital values to produce a converted image. The image conversion in Equation 15 is applied to a series of test images and a series of reference images to produce integrated test and reference images 1 and Ir, respectively. Test and Reference The images It and IR are then compared using well-known image processing techniques to produce a composite image. The composite image indicates the temperature difference between the test and reference images; the absence of the expected warm or cool area implies the presence of defects. Usually, the short_circuit will produce a higher-scale Wei, so it will be (4) relatively hot. The open circuit reduces the current and remains relatively cold, making it more difficult to image with the image. The enhanced thermal contrast provided by the present invention allows for a sensitive IR _ _ _ image, which can indicate the previous benefit

Defects of many types that are difficult or impossible to examine using conventional IR thermography. Such defects include '

What is a relatively cold defect artificial, which shows that the aforementioned implementations such as a line 1885 represent an open sign. Defect Location Interpretation ‘The current, and therefore the line temperature. These lines still appear in the composite image. Short circuits between the two lines increase the current flow through the lines. Thus, although there is no defect in itself, the part of the butterfly that is recorded in this 35 1303392 f is called a short circuit "defective person / dish degree. These features will appear in the synthetic shadow silly

Jin^ im - , became a "defective person =" of the road. The figure shows that the defect test of the _ line type associated with the open circuit includes the defect data associated with the space _ and the defect-free area of the work object. Like information. unfortunately

The difficulty Η 20 ' Although it is not based on the measured data, it can correctly display a synthetic image 2GGG, which shows the point-like crane defect 5, the linear surface defect toilet, and the corner shape defect 2015 appear in the synthetic image in. Defective artifacts obscure the location of the defect. This side—implementation side and New Zealand-_, shoots out defects from the artificial signs of defects to achieve the gamma, positioning and analysis of defects. The image processing that is distinguished from the artificial image of defects by its phase is 'locally', depending on how the artificial image of the defect is classified. The processing, for example, differs in the dot pattern, the line type, and the corner type defect (4). The age of the stomach material in the county is lack of gambling in the image analysis, resulting in the positioning of the fine ship _. According to the present invention, the image processing method can analyze the defective dragon and the defective person's money image to deal with this problem. (In the combination of image consumption _, - phase __ material, for the sake of brevity, it can be called a defect); similarly, the defect artificial sign data can also be called "defective artificial sign". "Defect" or "defective person reading", which refers to the physical characteristics of an imaged object, the physical image of the domain table: miscellaneous, which will be cleared in the domain circumference. Figure 19 is a composite image_ The cap shows three representative defects, namely - dot-type defect Cong, - linear type defect 1907, and a corner type defect. These defects are assumed to have the same size, but due to their respective Defective artificial signs (9) 5, 1920 and 1925 have different defect images. Unfortunately, the real synthetic Chinese images are not as follows: _ artificial county, miscellaneous work (S) 36 1303392 implementation of the present invention Examples include image processing techniques that can be used to search for artifacts of defects by type. Image processing techniques that classify artifacts by type include: pattern recognition and morphological analysis (m〇 Rph〇1〇 Gical analysis). The "Image and Video Processing" edited by A1 Bovik (2000) and edited by Ε·Dougherty and j Ast〇la (1999) (" "Non-linear Filters fOT Image. Processing")' describes the theory of image processing and mathematical morphology, which is also applicable to certain embodiments of the present invention. The two books are hereby incorporated by reference. Figure 21 is a flow chart showing a defect positioning/showing method 2100 according to an embodiment of the present invention. The Takizawa method 2100 accepts a synthetic image 21〇5, which is also optional. Grounding is filtered using a fast Fourier transform (FFT) low-pass filter (steps 21〇7), where the defect-poor material is surrounded by the defect artifacts. Subsequent processing is automatically performed in the defect artifacts. The defect data is located to generate a list of defect coordinates. In one embodiment, the composite image 21〇5 is an IR synthesis image of the aforementioned type; however, the image of the other types is also The defects surrounded by the artifacts of the defect to be located will be revealed. The interpretation according to the invention can be used to clamp such targets, such as 'images acquired by video optics and/or Objectives and target artificial signs revealed by the nucleon type experiment. The link between gray scale and binary morphology is described in the Bovik reference material mentioned in Lu Yueil's face, which can be applied to some embodiments of the present invention. Among them is to distinguish defects from their associated artificial signs. This link is based on an image J(x), which can be reconstructed from a combination of S-value limits. This reconstruction can be expressed in the following mathematical formula: ® V (^) = {x ED : l{x) > v\ — 00 < v < 00 (16) where Θν(/) is a threshold Recording the threshold of the grayscale synthetic image 7 of v, and 37 1303392

Dcif is the number field of the image /(x)=supveR{x€0v(/)} (17) The translation 2100 uses a similar type of image reconstruction to define the threshold level and optimize it' According to the improved defect positioning effect. A ^/1 car; MFA (morphological filtering algorithm) 2110, applies each of a number of threshold levels to the filtered composite image. The MFA 2120 is repeatedly executed for each threshold level, which is shown in the figure by wrapping the MFA 2110 in a for_1〇〇p 212〇A and 212〇B to generate a group/section. A synthetic image 2i25[lj] defined by the limit value. In each image 2125, all of the pixel values of the synthesized image after the wave is greater than or equal to the applied threshold level are all represented by a logic level (eg, logical 壹), and All those whose pixel values are less than the threshold level 2115 are all represented by the second logic level (for example, logic zero). In one embodiment, the threshold level 2115 is generated by the filter 21〇7, and the standard deviation of the pixel intensity values from a filtered composite image is the unit of representation. The histogram analysis is used to calculate the mean " and the standard deviation σ, the true threshold level is in 〃+ σ·:Γ, and the 忖 is σ is the single (four) input threshold Value level. The MFA 2110 will be described in detail below using FIG. ® The next for-loop (2130Α and 2130Β) processes each image 2125 with a second MFA 2135, which is type-specific. For example, in the case of linear artifacts, MFA 2135 removes artifacts from other types (such as corners and dot patterns), leaving only the linear shape. The For-loop 2130 produces a set of images, each having an artificial sign of a linear form of y. In the figure, the top side image 214 〇 [1] includes two line-shaped defects: the dot-like and corner-type defects of the image 2125 [1] are removed. The remaining images 2140 [2.·ζ·] may have more or fewer artificial signs. The MFA 2135 will be described in detail below using FIG. · 38 1303392 In a dying sequence of operations, a defect location interpretation 〇18 2145 can analyze the defect artifacts in image 2140 to accurately distinguish defect coordinates 215G from defect artifacts. The following is a detailed description of a version of the DLA 2145 that is specifically used for linear shape defects. Figure 22 shows an embodiment of the MFA 2110 of Figure 21 that is repeatedly applied to the composite image 2105 to produce a sequence of z• filtered images 2125. First, the threshold level 2115 is applied to the FFT filtered composite image (step 2·). - A morphological closure step 2205 (bribe after expansion) is applied to the results of the step leakage to add the secret to the Laiwuwei County, and the money-continuation-mineral treatment will be the noise or the defective side of the pixel. The resulting small image effect is removed (step 2210). MFA 2110 can thus produce one of the images. As shown in the figure, MFA2H0 is repeatedly executed for each of the threshold levels 2115. Fig. 23 is a flow paste in which an embodiment of the type specific alpha morph in Fig. 21 is shown, which is applicable to a linear type defect artificial sign. The job welcomes the post-wave image 2125 from the brain 2ιι〇, and uses a type-specific test that can rotate any artificial sign other than the linear type of artificial sign. Step 231 provides a specific morphology of the crane based on the position of the artificial sign on the image (4). For example, an artificial sign in the area where the filament is expected to appear in it can be read by any of the rest of the person, and in this case, several techniques can be utilized. Any one to fill it (step 2315). , the main =, _ from the artificial signs _ and _ wealth derived crane specific restrictions 2; 2 22 232G can filter the image from step 2315. A specific list of the types of the main ^, ', and _ artificial signs and ranges. For example, the linear type of artificial sign in the middle can be due to its circle type factor ("Line J is not true edge" linear type artificial sign must be close to vertical or horizontal), and complex (such as 'linear type The artificial sign of the state, unlike the corner type defect, is not filtered by the intersection of two adjacent boundaries of 39 1303392. The filtered image of the result is only included in the image 2140. Artificial signs of type. In one implementation!, '5 ^ « like 2140 has two pixel values: background (0) and artificial signs (1). The artificial signs appearing in the form of areas touching the pixels are all set to 1; the surrounding area ^ in the form of pixels is not fixed to 0. The execution details of each block in the MFA 2100 are familiar to the image processing. The artist is a well-known. As shown in Figure 21, MFA 2135 is for each The threshold value is repeated and executed. The entire set of morphological operations used in MFA 2135 is specifically specific to the linear shape in this example, but these operations can be modified if necessary. Conforms to, for example, the lack of point and corner patterns If only the artificial signs of the dot pattern are to be dealt with, then _ MFA 2135 can be woven to image the image, and the artificial image is larger than the specific maximum area, etc., which may be related to the line shape and other types. Removal of artificial signs of the state. If only the artificial signs of the corner type have to be processed, the MFA2135 can be adjusted to remove the smallest surface of the mosquitoes that are not represented by the _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Conventional image processing techniques are utilized to select such artificial and other types of artificial artifacts. Figure 24 shows an example of a defect-specific positioning interpretation (DLA) 2145 of FIG. 21 that utilizes an array of type-specific images.产生 产生 产生 产生 受 DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL DL In the second step, the linear defect filter!! LDF _ method produces a z-LDF architecture 2405[1··ι] with a car j, each image 214 There is an ldf wood structure in addition to the specific image 214〇 (10) The interpretation program also receives the synthetic image of the original city peak, and the value of the initial threshold value of the image measurement can be used to obtain the value of the threshold value of the threshold value of the continuous image. (S) 1303392 and the number of threshold values / (Because the 2140 is an image limited by the threshold, the number of thresholds and values is the same as the number of images 2140), etc. as input 〇 Figure 25 shows an LDF array 24〇5 according to an embodiment of the invention (the linked list array of Figure 2 is applied. In this example, four threshold levels and associated processing are applied to generate 24〇5[1] to 24〇5(4) Four LDF structures, although the number of arrays in the real world may be more or less. These arrays are adaptable and can be applied to the linear shape of the human body, but can also be modified to apply to other types. Each of the LDF architecture feathers includes the following features. 1· - Threshold field 25GG, which can store finer than synthetic images to produce threshold levels on each binary 2140[/]. 2 - A number of fields 2505, which can store the number of identified defects w, which is two in the actual example of the figure %. 3· — Image field 2510, which can store each binary image 2i4〇[/]. 4. Defect architecture 2520 [1·/|-array 2515, where] is the number of defective artifacts in each image 214〇|7]. Each defect structure 252 〇 stores the top of each linear shape defect artificial sign, or "peak", the χΑ γ coordinate of the pixel (the defect is assumed to be located near the top of the artificial sign of the linear defect). Each defect architecture further includes a pixel-valued field 2525 that stores X and γ coordinates corresponding to the peak pixels in the composite image 2105. φ 5. The linear structure 2535 [1·/|-array 253〇, each associated with the defective artificial artifact in the corresponding filtered image 21 side. Each linear structure 2535 includes an area field 2540 that can store the area of the defect artificial sign (in pixels) a rectangular field button, such as (6) Qiao, which can store the artificial image of the defect - rectangular Left, top, right, and base marks; a belongss-to-feld 2550 that stores a line index that can associate a linear type of artificial sign with a feature of an image, the image being lower The threshold value ("" is not obtained when there is such a line; and a field 41 1303392 (contains fidd) 2555 'which can store the _ line shape in the filtered image; the type defect is higher In another post-wave image obtained from the threshold level:: The line index of an array associated with one or more related lines. Back to Figure 24, an LDF architecture 2 is analyzed. 2410) to obtain the first-f LDF gradient peak proflle 2415 [w], each of which is characterized by a specific defect in each image. The following discussion of Figures 26, 27a-27d and 28 will illustrate the processing of the value side view 241. Figure 26 shows a similarity to the step of Figure 21, which illustrates the nature of the image 2600. The image 2_ includes a pair of vertical linear defects, artificial signs 2 and A2, and a point-like artificial sign A3. The edges of the defective artificial signs are pasted to reveal a lack of a clear image boundary between the defects and the artifacts. However, it is assumed here that the defect associated with the artificial sign of the linear type defect is close to the top 'or 'peak' of the artificial sign and the defect of the side of the point type is concentrated in the associated artificial sign. Figures 27A-27D show binary, pattern-specific images 27〇5, 2710, and 2715 similar to the image of Figure η. According to MFA 211, each image represents a synthetic image 2600 with different threshold values; the dot pattern artificial sign t A3 is removed from each binary image after subsequent application of the specific MFA 2135. Each image is stored in field 2410 of LDF architecture 2405 (Fig. 25) in conjunction with other image and artifact identification information previously described in relation to Fig. 25. The image 27〇〇 (Fig. 27) includes a pair of linear artificial signs LA1 and (1). A related indication_2720 touches the artificial signs LA1 and LA2 as associated with the artificial signs A1 A A2 in Figure %, respectively. As the threshold level increases, the artificial signs of the linear pattern become smaller and thinner, and may disappear or split into lines that are not continuous. For example, in ® 27A it appears as a single line of defect artifacts la2, in Tujia

42 1303392 becomes a pair of linear artificial signs LB2 and LB3. A marker 2725 identifies the artificial sign LB1 as the artificial sign LA1 belonging to the image 2700, and the artificial signs LB2 and LB3: belongs to the artificial sign LA2 of the image 2700. The "ownership" of this artificial sign is recorded in the "home" field 255 of the LDF array associated with image 2705; likewise, the "inclusion" field 2555 of the LDF array associated with image 2700 records the artificial sign* LA1 "includes" the artificial sign LB1, while the artificial sign LA2 contains the artificial signs - LB2 and LB3. Figures 27C and 27D respectively include different combinations of defective artificial signs and corresponding indicia 273〇 and 2735, which reveal the "belonging" relationship between artificial signs in different images. Figure 28 shows a tree listing 2800 which is a collation of the information in Figures 27B-D, 2725, 2730, and 2735. The tree-like representation of 28〇〇 shows the relationship between the artificial signs of defects A1 and A2 and the linear type artificial signs in the binary image obtained by using the larger threshold level. For example, the artificial image LD2 of the image 2715 belongs to the artificial sign LC3 of the image 2710, and the latter belongs to the artificial sign LB3 of the image 2705, which in turn belongs to the artificial sign LA2 of the image 2700. Similarly, the artificial image LD1 of the image 2715 belongs to the artificial sign LC1 of the image 2710, and the latter belongs to the artificial sign LB1 of the image 2705, which in turn belongs to the artificial sign LA1 of the image 2700. Returning briefly to Figure 24, the LDF gradient peak side view step 241 uses the data in Figures 27a-27d to generate y_peak side views 2415 [W], each of which has a side view. Figure 29 is a peak side view of the nature of the nature of Figure 29, showing the relationship between the four defect artificial signs 29〇5,291〇, 2915 and 292〇 all associated with a common defect artifact. Side view 2900 shows the relationship between the threshold level and the position of the peak pixels in each of the defect artifacts along the y image axis. The pixel intensity is diced along a whole pixel, which is a curve 2925 parallel to the artificial image of the gradation of the linear shape defect of a gray scale image, showing how the pixel intensity generally increases as a result of encountering the artifact of the defect. The artificial sign 29〇5, measured, 拠 and measured representative curve 2925 fine 43 1303392 each shows the different threshold value level of the threshold obtained. The y position of the peak (top) pixel of the upper "+ defect artificial sign" returns to Figure 24. The positioning step 2420 extracts the data of the touch pattern of Figure 29 from the LDF array to produce a number of two-dimensional arrays. The number of artificial signs in the image of the LDF architecture 24〇5(1) is the size of the image of the LDF architecture. The two arrays are as follows (the official definition is given by the pseudo-code below): 1· y[j,i] is the y coordinate of the defect structure 252〇ΰ] of the LDF structure, which can indicate

The defect related to the j-th personal sign of the image 2510 with the threshold limit imposed by the level i; 2· LineD, i] is the index of the line containing the defect j below the margin bit (the interpretation of the pseudo-code) How the index should be selected); and 3· dy[j5i] is the difference between the 乂 values from the defects 252 of the LDF architecture 24〇5[1+1] and 24〇5[1]. Step 2715 performs the following initialization and loop procedures for each artificial sign: • Initialization: (each j=1,...,Dimj):

Line[j5l]=j; y [j,1 ]=LDF—Array [ 1 ] .Defects [j] .y At the threshold value #1 (initial threshold), all filtered lines are It is assumed that there is a defect, so Line[j,l] is the index of the line and is the assumed y position of the defect associated with this line. Repeat i: i=l,...,DIMi_l (each j) 44 1303392 /* i-threshold index; j-index of the line 2535 from LDF structure 2405 [1] corresponding to the lowest threshold level. */ 1· Kp corpse argmin " LDF Array[i+1]·

Defects[ LDF_Array[i]Xines[Line[j?i]].Contains[k]].y? } 2· Line[j,i+1]=LDF—Array[i].Lines[Line[j,i ]]·

Contains [kopt] 3 · y D 4+1 ]=LDF_Array [i+1 J.Defects [Line[j 5i+1 ]].y 4· dyD^yD.i+iJ-yD^] Step 1 above And 2, for the selected line[i,j] of the threshold level/, the line of the next threshold level (i+Ι) belongs to Hneiy]. Among the lines belonging to the threshold level (i+Ι) of line[M], the line with the lowest y value is selected as the line most likely to contain defects (the y-axis of each image is from top to bottom) Therefore, the lowest 7 value represents the highest point). After the line Line[j5i] is selected with the lowest y value, the y value of the defect associated with this line is taken out of the corresponding line architecture. The value 咧, 1] defined in step 4 of the immersive code listed above is applied in the following manner to define the LDF gradient. As shown in Figure 28, when the line defect has been displayed using a tree-like list of artificial signs, and the peak side view of the LOT gradient has been calculated, the end of the line associated with the defect (according to the previous assumption that the upper end is Defined as the maximum value of the LDF gradient. The LDF gradient for each - threshold value index z is defined by (18) (18) 1303392 LDFGradh = ^Th where y. is the defect index '/ is the threshold level The index, and ^ is the interval between the threshold levels. DTh is a constant, so there is the following formula: max, LDFGrad[ ij] = _ DTh ((9) - additionally 'by and only applied to the ldf gradient maxima Position, the search for the maximum value of the LDF gradient is equivalent to the search for the minimum value of the column in the column as calculated by the LDF_peak side graph. In many cases, due to the image bit φ The pixel's 娄i: measured in the direction of 'the dy's minimum value is zero, and below this granular level, the defect's y position may be the same as the adjacent threshold level. When the minimum value of dy will be set, the LDF is slightly maximal (minimum value, such as time, the multiplicity of position will cause The need for the calculation of the first and last indices of the minima fraction in the column of [] at the array. The data structure of the MaxLDFGrad type contains the following components: 1· is the threshold of the maximum value LDF gradient 2· Ih.Ind· is the index of the threshold level; 3. XJne Ind· is the clue of the defect related to this threshold level; 4· is the minimum of the array edge ^[] Value; and 5. X and y are coordinates of the defect location. Step 2420 uses the analysis of the LDF gradient peak side view 2415 to locate the defect associated with the linear defect artificial artifact specified in the line architecture of the LDF array. Step 2420 performs the calculation based on the Hi and L^imaxLDFGrad architectures (input data 2425) associated with the _ and the last indices of the minima fractions in the array of arrays: 46 1303392 - or two architectures are used to guide (4) The position of the defect to be related. This position can be defined as the defect structure 252〇_(xy) from the LDF architecture 24_, where [] is obtained from one of the two or the LDFGrad. In the case, the missing position is in the middle of the two. The location list can be accurately positioned among the linear artifacts of the image. The final step 2435 uses the reference positioning marker 244〇 to replace the image coordinates (10) f with the actual defect coordinates 2150 (Fig. 2丨). For this purpose, the object being tested is assumed to be positive with each _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 'But if necessary, these operations can be easily modified to apply, for example, to defects in point and corner patterns. The defect location interpretation method 21 (Fig. 21) can therefore locate various types of defects in its corresponding artificial signs. Although the present invention has been implemented as a practical complement, it is apparent that the various changes in the actual complement are apparent to those skilled in the art. The interpretation method disclosed by the present invention is a body-watcher because it does not rely on the image crane (video, infrared, spectrum type, scanning type, material) and does not rely on image detection H. As a result of this generalization, the same approach can be applied to computer-based inspection secrets for many different types of cameras and side n's and for the A-domain. The proposed autofocus system can utilize the imaging hardware of the existing imaging system, which does not require any additional components; with the tilting of the present invention, the image can be analyzed and an appropriate command can be generated to focus the image. In addition, as described above, the panning stage can also be used without the need to achieve autofocus, but can control the adjustable objects. Thermal image capture, artificial image detection, and autofocus methods and systems can also be applied to many forms of electrical circuits, including integrated circuits, printed circuit boards, microelectromechanical systems (MEMS), semiconductor wafers, and certain Health doctor samples, etc. In addition, although the IR inspection system described in the present invention employs an integration of electrical excitation with fine imaging that produces artifacts of thermal defects, other forms of inspection systems of the present invention

47 1303392 Other methods of excitation and/or imaging are used to generate other types of defect artifacts. For example, some imaging systems require the use of white light images, integrated spectrum, or magnetic fields. The present invention is directed to embodiments described in which defects are isolated from defective artificial signs and is not limited to the IR inspection system of the type described therein. Therefore, the spirit and scope of the appended claims should not be limited to the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 (Prior Art) shows that it can be applied to a part of an active panel touch on an LCD panel. Figure 2 (Prior Art) shows details of a portion of a conventional pixel 11 that is used herein to reveal possible defects. Figure 3 shows an inspection system 3 in accordance with an embodiment of the present invention. Figures 4A and 4B are a pair of sample images and 4〇5, which are used to illustrate the method by which certain embodiments are based. Fig. 5A and 5B are the scale diagrams of the Z-spot and the 4()5±-mumrace traces. 6A (four) #'s _ added to f 彡 like tearing and milk on - silky riding Qi Qing, its focal length measuring tank as a function of the Z position curve. Figures 7A and 7B are graphs of the difference in the focal length measurement as a function of Z position, which is applied to the image and the side. ^ 8A money is applied to the image and the milk - the nonlinear gradient is just ρ, and its focal length is a function of the 2 position. _ and (4) are different from their parameters and secrets at the time of 8 o'clock = non-linear gradient FMF, material_value as a map of & (four) function.

For the self-optimized rice--the automatic focus interpretation method 1000. W 48 < S ) 1303392 Figure 11 shows a flow chart 1100 which is based on gradients and is adaptive, one embodiment of true time autofocus. Figures 12A through 12D are graphs of four gradient-based autofocus interpretations, 'focal distance measurements FM and velocity V, relative to time in seconds, respectively, applied to the same image. Figure 13 shows the effect of the aforementioned interpretation on maintaining the focus during the movement of the object in the XY plane. Figure 14 shows the dynamic response of the FMF applied to the autofocus of Figure 13. Figure 15 shows a test system 1500 comprising a conventional board 15〇5 and an inspection system 1510 according to an embodiment of the invention. Figure 16 shows a board 16 which enhances testability in accordance with an embodiment of the present invention. Part of the cockroach. Figure 17 shows a portion of an LCD panel 1700 in accordance with another embodiment of the present invention. Figure 18A is a graph 1800 showing a thermal reaction of a sample defect and its surrounding area. The graph 1830 of Fig. 18B shows the test vector and image acquisition time control for enhanced thermal contrast between the defect and its surrounding area during the winter. Figure 18C is an experimentally obtained image 1880 showing how the foregoing embodiments can produce sufficient thermal contrast to indicate an open circuit. Figure 19 shows a composite image 1900' showing three representative defects, a one-point type defect 1905, a one-line type defect 1907, and a corner type defect 19 〇. Figure 20 is a pre-defined composite image 2000 showing how a one-point pattern defect 2005, a one-line shape defect 2010, and a corner pattern defect 2015 can appear in a composite image. Figure 21 is a flow chart showing a defect location interpretation 2100 in accordance with one embodiment of the present invention. Figure 22 shows an embodiment of MFA 2110 in Figure 21, which is applied repeatedly to a composite (S) 49 1303392 image 2105 to produce a sequence of z• filtered images 2125. Figure 23 is a flow chart showing an embodiment of the type sMFA 2135 of Figure 21 which is applicable to linear type artificial signs of defects. Figure 24 shows an embodiment of the Defect Location Translation (DLA) 2145 of Figure 21. Figure 25 shows an iLDF array construction 24〇5 in accordance with an embodiment of the present invention (Fig. 26 is similarly contemplated to be a filtered composite image 2600 of a nature similar to the system 21〇7 of Fig. 21).

Figures 27A-27D show binary, type-specific images 2705, 2710, and 2715 similar to the image 2l40 [l"i] of Figure 21. Figure 28 shows a tree listing 2800, which is a collation of the information in Figure 27, ugly-〇, 2725, 2730, and 2735. Figure 29 is a perspective view of the nature of the peak side view of Figure 29, which shows the four major defect artifacts associated with a common defect image 2905, 291 〇, 2915 and 292 【 [main symbol description] 110 pixel 125 gate line (control line) 212 common line

1〇〇 active board 115 source line 200 transistors 215, 216, 220, 105 first shorting bar 120 second shorting bar 210 capacitance 225, 226 short circuit 227, 228, 229, 230, 232, 233, 235 open circuit 305 image Detector 315 object 330 motion controller 360 motion control module 1 〇〇〇 auto-focus interpretation 1040, 1043, 1045, 1 〇 55 step 300 inspection system 310 video camera 320 325 box grabber 335, 340 Ζ 355 image analysis module 400, 405 sample image 600, 700 flat area 1005, 1010, 1015, 1020, 1025, 1035, 1030 curve 1050 position 1100 flow chart 1105 data 50 (S) 1303392 1110, 1120, 1130 1,140, 1150, 1160, 1170, 1180, 1300, 1305, 1310 Image 1500 Test System 1505 Board 1510 Inspection System 1512 Shorting Bar 1515 Detector 1520 Computer 1525 Frame Grabber 1530 Signal Generator 1600 Board 1605 Pixels 1610 Source Line 1615 gate line 1620 common line 1625 source rod 1630 common rod 1635 gate rod 1700 board 1705 pixels 1710 source line 1715 gate line 1720 common line 1725 source rod 1730 gate rod 173 5 common rod 1740 area 1800 curve 1805 Figure 1810, 1815 area 1820 reaction curve 1830 curve 1840 reaction curve 1845, 1850 window 1855 high side part 1880 image 1885 line 1900 synthetic image 1905 point shape defect 1907 line shape defect 1910 corner type defect 1915, 1920, 1925 defect artificial sign 2000 synthetic image 2005 point shape defect 2010 linear shape defect 2015 corner type defect 2100 interpretation 2105 synthetic image 2107 step 2110 interpretation method 2120A, 2120B, 2130A 2130B for-loop 2115 threshold value 2125 image 2135 filter interpretation 2140 image 2145 defect location interpretation 2150 defect coordinates 2200, 2205, 2210, 2305 2310, 2315, 2320, > 2410, 2420, 2435 2325 limit 2400 interpretation 2405 architecture 2408 input 2415 gradient peak side view 2425 data 2430 defect location list 2440 coordinates 2500 threshold field 2505 number field 2510 image field 2515, 2530 array 2520 defect architecture 2525 pixel value field 2535 linear architecture 2540 Area field 2545 rectangular field 2550 home field 2555 contains fields 2600, 2700, 2705, .2710, 2715 images

51 1303392 2720, 2725, 2735 mark 2730 defect artificial signs 2800 tree list 2900 peak side view 2905, 2910, 2915, 2920 defect artificial signs 2925 curve

< S ) 52

Claims (1)

1303392 X. Patent application scope: 1. A usable-image test method includes: a method of autofocusing on a piece to find out its defect, the party a· obtaining a b. The range includes - the best focus surface; _ image 'the focus surface cd Z image is applied - the first focal length measurement function is simple - the first plurality of focal length analysis first - complex _ focal length measurement; And a plurality of obtained - the second has a number of t = 帛 - lion coffee measured value including fixed focal length flap value such as ▲ face - touch object _ slave analysis including judgment 3 · as claimed in the scope of patent application, i 兮〃 For the touch surface _ „ 即 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦 焦The judgment includes the width of the sensitivity curve. 'Curve 5> The method of claim 3, wherein the determination of the curve is related to the sharpness, and the determination of the shape of the curve includes the measurement of the sharpness. If the method of applying for the scope of the patent item is included, it further includes A method for measuring a second plurality of focal lengths. 7. The method of claim </ RTI> wherein the first plurality of focal length measurements are plotted against a range of focus planes. The first curve, and the second plurality of focal length summer measurements are drawn as a second curve when plotted against the range of the focus surface, and the method further comprises analyzing the first and second curves. 53 1303392 8· For example, the method of claim 7 is a bee value, and the method further includes adjusting the focus surface to the best focus corresponding to the best focus surface. 9 As in the patent application range δ, the focus on the curve The peak value of the only peak Γ in the face is the peak of the best focus surface is -10. The method according to claim 9 of the patent application, wherein there are two second curve peaks. ^ In the range of the focus surface, as in the patent application The method of the range ^, wherein the step of the range of images comprises: adding the image side device relative to the object. ',,, and obtaining a complex number in the face 12. - an image detector can be included in the method Have ·· auto focus to find out its lack In the method of trapping, the party i obtains the first image of the object on a first focusing surface; the first image-to-image volume-_quantity —^-focal length measurement value. [the first-focus surface adjustment speed by the _ The focusing surface moves to the _th surface. The second image of the object is acquired on the second focusing surface. The second image is ..., and the focal length measurement function is applied to the second image to obtain the second focal length. The second focal length measurement is compared; and == fruit is moved by the second focus plane to a second focus plane to -=· as in the method of claim 12, wherein the comparison includes the first and the The measured value in the two focal lengths is different from the measured value gradient. ', ,, M · As for the method of applying for the patent item 13, the (four) - the focus surface to the second pair of the opposite side of the face Occurs in the second direction. A method of claim 14, wherein the first image is an infrared image. 11. 111. IV. V. VI. VII. 54 13〇3392 6 · Can be - image predator acquisition, - auto focus system, the system has a good focus (four) for finding out - defects on the object h a focusing mechanism, can be installed in 4, like to detect the best difficult face - focus adjustment within the focus ii. a frame grabber 4 pairs of machine ritual fine - focus adjustment speed to adjust the focal length; image; The mechanism captures a plurality of focal lengths when adjusting the focal length within the focal length range. · · · 111. An image analysis calculates a number of errors to calculate a focal length measurement value; clothing - for mother - image to ιν 纟 coffee for 4 meals _ - focal length measurement Ϊ́λ changes the focal length adjustment speed as the focal length gradient. 18: 1:: The system of item 16, wherein the focusing mechanism comprises a translation stage. on. 4The system of the 17th item of Liganwei, in which the image detector is installed on the translation stage 1=^^利细16-shaped green, the cap image analysis mode_speed (four) module, the system uses a general purpose Computer to implement. The number of sub-tests 1 at least one of the function wipes is used to calculate the focal length measurement. 2ι.- An automated method for finding at least a defect on a target object, wherein the object includes at least a defect area associated with the defect and a defect-free area if there is no defect, the method comprising: a Acquiring a plurality of images of the object in a range of focus surfaces, wherein the range of the focus surfaces includes an optimal focus surface; the amount of focal length b is a first focus measurement function is applied to the image Get a first number! Measured; C· analyzes the first plurality of focal length measurements; 55 1303392 de According to the analysis, several focal length measurements are applied to the images; the diop-focus setting function is obtained to obtain - the second complex number __ 刺面面 = includes spatially associated image data associated with the object Two defects related to the defect area; the county artificial signage associated with some parts of the workroom; and the analysis of the artificial image of the defect. The method of claim 21, wherein the first plurality of focal length measurements comprise a piano value 'and wherein the analysis of the first plurality of focal length measurements comprises determining a peak value such as number. f. 23. If the second sergeant of the patent application scope, the knives of only a few knives, and the number of focal lengths of the priests, are displayed in the range of the focus surface, a curve is drawn, and the first plurality The analysis of the angular distance measurements includes determining the shape of the curve. [24] The method of claim 23, wherein the curve is related to the width, and wherein the determination of the shape of the curve includes the width of the determination curve. 八25·If the method of claim 21 is included, the method further includes The defect artificial sign data is classified into one of a plurality of individual physical signs. 26_ The method of claim 25, wherein the pattern of the artificial sign comprises at least one of a dot pattern, a line pattern, and a corner pattern. 27. The method of claim 21, further comprising filtering the image data to remove at least one type of defect, thereby producing a type of image specific. 28. The method of claim 21, further comprising: a. applying a first threshold to the image data to generate a first threshold-limited image, the first threshold defining the image in the defect region The first defect area and the first defect area are covered by the 56 1303392 and the first defect artificial area is extended into the defect-free area; and b· is applied to the image data - the second threshold To generate a second threshold value defining image, the second threshold value defining an image in the defect area to trace a second defect area and a second defect covering the second defect area and extending into the defect free area Artificial signage area. 29. The method of claim 28, further comprising selecting a first peak from the first defect artificial image region, selecting a second peak from the second defective human image region, and calculating the first and the The gradient between the two peaks. ^ 30. A system for positioning a defect on an X-test object, the system comprising: a·- an excitation source 'applying a defect-test vector, wherein the defect exhibits a reaction to the applied test vector; b•- f 嫌 II II, its location is set to the image of the side defects, these images include defect data and defect artificial signs data covering the defect data; C · - Μ ' ' ─ ─ cover the best difficult face 1 _ Adjusting the focal length of the image detector, the focusing mechanism adjusts the focal length with a focal length adjustment speed; d. A frame grabs, when the money is constructed in f, the range of the range _ the focal length _ catches a plurality of images; e · - image The analysis module calculates a plurality of focal length measurement values, and calculates at least one focal length measurement value for each image; and the f·-speed control group, calculates a “focus distance measurement value” from a plurality of distance-valued towels. The gradient 'reacts in response to a difficult amount of focal length and _ focal length adjustment speed. Shi Shenming specializes in her third system, including an image processor that analyzes defects, artificial signs, and defect data to identify defect data in defective artificial signs. &lt;S) 57 1303392 Supplementary amendments to the scope of patents No scribes No. 0931379854 No. September 1997 Revision of the month of the month (32. For the system of claim 31, the analysis of the defect artificial signage data includes the image Applying a plurality of thresholds to generate a plurality of threshold-limited images, wherein the plurality of threshold-limited images include corresponding defects associated with the defective area 33. According to the type, the defect image of each image of the image towel can be defined according to the type to produce a specific image of the type of defect artificial image. 〜 34 · The system of claim 33, wherein the image The processor includes a data structure to visualize different crane-specific images and types of defects. 35. The system of claim 34, wherein the image processor can be established at the peak 1 of the symbol. Gradient., αΑ 36. The system of claim 35, wherein the image processor utilizes the position of the trap. 37. For the system of claim 3 (), the reverse is - —Reaction. 3—8. For the system of claim 30, wherein the image analysis module supports the reconciliation test function and the at least one of the focal length measurement functions to calculate the focal length measurement. 39. An automated method for identifying defects on a circuit, the method comprising: a. focusing on a circuit, the focusing includes: i. acquiring a plurality of images from a range of focus planes The range of the focus surfaces includes an optimal focus surface; the image-to-focus measurement function is used to obtain a first plurality of focal length measurements; 1U. analyzing the first plurality of Focal length measurement; 1V· Applying a second focal length function to the images according to the analysis to obtain a second plurality of focal length measurements; 58 1303392 b. Applying a defect to the defect via the circuit at a first time The test vector, wherein the defect exhibits a test image of the defect at a second moment in which the test is directed to the heat reaction, and the simple reaction-like scale constant is separated from the first time by less than three time constants. 40. If applying The method of item 39 of the benefit range is separated by two time constants. 1. The method of item 39 of the scope of the patent application is a time constant phase separation. The second moment is the first moment and the first moment is less than one of the younger ones. And the first moment is less than 41, as in the method of claim 39, and further comprises repeating steps (9) and (6) to capture a plurality of test images at least twice. 43. A reference image of the method of claim 42 of the patent application. Further comprising capturing between the plurality of test images 44. A test system comprising: a_circuit' having at least a defect located in a defect region, the defect region having an initial temperature; b· - Infrared image fine n, its position setting can accept infrared beam from the defect area; C · can obtain an optimal focus surface for the image detector autofocus system, the auto focus system includes: Both ", and stack" can adjust the focal length of the image in one of the best focal planes, and the focus mechanism adjusts the focal length with a focal length adjustment; The mechanism captures a plurality of images when the focal length is adjusted within the focal length; (S) 59 1303392 111. An image analysis module calculates a plurality of focal length measurements, which calculates at least one focal length measurement for each image; IV The speed control module calculates a focal length measurement gradient from a plurality of focal length measurements, and changes the focal length adjustment speed in response to the focal length measurement gradient; and the d. At least one test signal output of the circuit, the signal generator applying a test vector to the defect via the circuit, wherein the defect exhibits a thermal reaction to the applied test vector; e· The medium image detector captures one of the defects before the test vector is applied and before the defect reaches 95% of the difference between the initial temperature and a final equilibrium temperature. 45. The test system and system of claim 44, The image debt detector captures images before the gap reaches approximately 86.5% of the difference. 46. For the test system of claim 44, the image detector captures images before the gap reaches approximately 63.5% of the difference. The test system of claim 44, wherein the signal generator comprises a second test signal output, the circuit comprising a first current processing terminal connected to the test signal output, connected to the second test signal output A control terminal of the terminal and a transistor of a second current processing terminal. 曰曰48. The test system of claim 47, wherein the test vector is an ac signal. 49. a test system, wherein the signal generator applies a second test vector to the control terminal via the second test signal output. 5〇· as claimed in claim 49 The test system, wherein the second test vector turns on the transistor. 51. The test system of claim 47, wherein the signal generator comprises - a third test signal output end - the test system further comprises a connection to the transistor a capacitor between the second current processing terminal and the third test signal output terminal.