JP2006337152A - Method and apparatus for forming teaching image, method and apparatus for forming image processing algorithm, image inspecting method, image inspecting device, program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a teaching image forming method, capable of forming a teaching image enabling algorithm to perform high speed processing so as to acquire processing becoming a target, in a self-developable manner in a teaching process, a teaching image forming apparatus, a program, a recording medium, an image processing algorithm forming method capable of forming proper image processing algorithm at a high speed, an image processing algorithm forming apparatus, an image inspecting method capable of properly determining the quality of an inspection target image, an image inspecting device, a program and a recording medium. <P>SOLUTION: The feature value of the inspection region image in an indicated inspection region is calculated for a teaching image, having an inspection region and a novel flaw adding and teaching image having a change inspection region image changed in the feature value of the inspection region image is formed, at a position where the calculated image is arranged. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a teaching image generation method and apparatus, an image processing algorithm generation method and apparatus, an image inspection method and apparatus, a program, and a recording medium, and more particularly to a teaching image generation method for extracting a desired region from image data. The present invention relates to an image processing algorithm generation method and apparatus, an image inspection method and apparatus, a program, and a recording medium.

  In recent years, application fields of image processing are expanding due to lower prices of image processing devices such as CCDs and image processing devices such as personal computers. In particular, the application of image processing to defect inspection in the industrial field is widespread, increasing the need for product and device yield, quality improvements and cost reductions by reducing visual inspectors. Therefore, at present, a wide variety of defect inspection apparatuses to which image processing technology is applied have been developed for defect inspection.

  In such an image processing type defect inspection apparatus, since there are various types of defects depending on the target product or device, it is common for engineers to design image processing algorithms and defect classification algorithms exclusively for each defect target. It is. However, the dedicated design of the image processing type defect inspection apparatus often relies on the experience of engineers, and requires a very large number of man-hours for its development and introduction.

  In recent years, an image processing type defect inspection apparatus having a learning function has been developed for such a problem. Learning algorithms used in the image processing type defect inspection apparatus include optimization methods such as neural networks and evolutionary methods.

  Japanese Patent No. 3476913 (Patent Document 1) (hereinafter also referred to as Conventional Technology A) discloses a technology (hereinafter also referred to as Conventional Technology A) relating to a learning type defect type determination apparatus to which a neural network is applied. Has been. In the apparatus of prior art A, first, before learning, defects are extracted by performing image processing on an input image that is a learning teaching image, and defect information (area, barycentric position, ferret diameter, The brightness of the defect is detected.

  Further, in the apparatus of the prior art A, defect information of the extracted defect is assigned to the input layer of the neural network that is a learning device, and the defect type (scratches, dust, exposure failure, development failure, etc.) is assigned to the output layer. Type). That is, in the apparatus of the prior art A, learning is performed by optimizing the intermediate layer of the neural network so that the correspondence between the input defect information and the output defect type is appropriate.

  However, the learning type defect type determination device of the prior art A is specialized for determining the defect type of the teaching image for learning, and the accuracy is high even when determining the defect type for an inspection target image that is slightly different from the teaching image for learning. It is known that an overlearning state is likely to occur that is significantly reduced.

  As a countermeasure technique for such a problem, there is a technique of an image processing program synthesizing method disclosed in Japanese Patent Laid-Open No. 11-282822 (Patent Document 2) (hereinafter also referred to as Conventional Technology B). In the conventional technique B, an image processing algorithm is automatically generated using a genetic algorithm which is one of evolutionary techniques. Further, in the conventional technique B, the defect type determination accuracy of the inspection target image similar to the teaching image is improved by adding the image subjected to image conversion as a new teaching image to the teaching image used for automatic generation. Has been reported.

  Japanese Patent Application Laid-Open No. 5-149728 (Patent Document 3) teaches a sample image obtained by imaging a non-defective sample by synthesizing one selected from defective patterns stored in an image memory. A technique for creating a commercial image (hereinafter also referred to as Conventional Technology C) is disclosed. In the prior art C, an operator inputs information such as the position, rotation angle, magnification, density, and color of a defective pattern using an input device while viewing a composite image of a non-defective sample and a defective pattern displayed on a video monitor. By doing so, an image for teaching is created.

Furthermore, as a method for creating the teaching image disclosed in the prior art C, an operator uses, for example, a mouse or a light pen as an input device for a sample image obtained by imaging a non-defective sample. There is a way to edit images directly with a drawing tool.
Japanese Patent No. 3476913 JP-A-11-282822 Japanese Patent Laid-Open No. 5-149728

  As described above, the conventional technique A is specialized in the determination of the defect type of the teaching image for learning, and the over-learning that accuracy is remarkably lowered even in the defect type determination for the inspection target image that is slightly different from the teaching image. There is a problem that it tends to become a state.

  In the prior art B, an image obtained by performing image conversion on the teaching image is added as a new teaching image. Specifically, the image conversion includes brightness change, contrast change, geometric change, and the like.

  Here, for example, a description will be given of adding a new teaching image created by changing the brightness of the teaching image in consideration of illumination variation in an actual environment. Actually, the brightness is changed by adding the density of each pixel of the teaching image by a fixed offset value α.

  For example, in the case where it is desired to improve the defect type determination accuracy for three cases where α is “5”, “10”, and “20”, the first new instruction for adding a density value “5” to each pixel is used. Three teaching images of an image, a second new teaching image 2 in which a density value “10” is added to each pixel, and a third new teaching image in which a density value “20” is added to each pixel Need to be added. That is, the number of new teaching images to be added increases according to the type of change amount after image conversion.

  Furthermore, even if attention is focused on a contrast change other than a brightness change, the number of new teaching images to be added increases according to the type of change amount. This means that it is necessary to add a number of new teaching images corresponding to the product of the type of image conversion and the type of change amount of each image conversion.

  As the number of all teaching images including the teaching images to be added increases, the time required for learning the image processing algorithm increases.

  Prior art C adds a teaching image created by synthesizing one selected from defective patterns stored in an image memory to a sample image obtained by imaging a non-defective sample as a new teaching image. Is the method. Therefore, since it is necessary to add new teaching images as many as the number of defective patterns to be added, the time required for learning the image processing algorithm increases as the number of all teaching images including the added teaching images increases. There is a problem that increases.

  Furthermore, in the prior art C, an operator uses the input device while viewing a composite image of a non-defective sample and a defective pattern displayed on a video monitor, and the like, the position, rotation angle, magnification, density, color, etc. Therefore, the burden on the worker for creating one teaching image is large, and the time and cost are increased.

  Another teaching image creation method disclosed in the prior art C is that a worker uses, for example, a mouse or a light pen as an input device for a sample image obtained by imaging a non-defective sample. It is a method of directly editing an image with a drawing tool. That is, since the method is artificial, there is a problem that a difference in contrast with an actual defect becomes large, and the reliability of what is obtained as a learning result may be reduced.

  The present invention has been made in order to solve the above-mentioned problems, and its purpose is for teaching which can cause an algorithm that learns a self-developed target process in a teaching process to be processed at high speed. It is an object to provide a teaching image generation method and apparatus capable of generating an image.

  Another object of the present invention is to provide a program that can generate a teaching image that can be processed at high speed by an algorithm that learns a target process in a self-developed manner in the teaching process, and a record that records the program. To provide a medium.

  Still another object of the present invention is to provide an image processing algorithm generation method and apparatus capable of generating an excellent image processing algorithm at high speed.

  Still another object of the present invention is to provide a program capable of generating an excellent image processing algorithm at high speed and a recording medium recording the program.

  Still another object of the present invention is to provide an image inspection method and apparatus capable of appropriately determining the quality of an inspection target image.

  Still another object of the present invention is to provide a program capable of appropriately determining the quality of an inspection target image and a recording medium on which the program is recorded.

  In order to solve the above-described problems, according to one aspect of the present invention, for teaching to generate a teaching image having an inspection region for causing an algorithm that learns a target process in a self-developed manner in the teaching process. An image generation method includes a step of inputting a teaching image having an inspection region, a step of designating an inspection region, a step of calculating a feature amount of the inspection region image in the designated inspection region, and a teaching image. A step of calculating the position where the image is arranged, a step of generating a changed inspection region image in which the feature amount of the inspection region image is changed, and a new defect addition in which the changed inspection region image is arranged at the position where the calculated image is arranged Generating a teaching image.

  Preferably, the predetermined end condition further includes a step of receiving an instruction from the user, and the step of specifying the inspection region specifies the inspection region based on the received user instruction.

Preferably, the position where the image is arranged is a position that does not overlap the inspection region image.
Preferably, the step of calculating the feature amount of the inspection region image includes a step of calculating a center of gravity position of the inspection region image, and the step of calculating the position where the image is arranged includes calculating the position where the image is arranged A search is performed from the vicinity with the center of gravity of the region image as a center, and the calculation is performed.

  Preferably, the method further includes a step of determining whether or not the inspection region is a region to be extracted, and the step of generating the change inspection region image is performed when the inspection region is determined to be the region to be extracted. A change inspection area image in which the area of the image is reduced is generated.

  Preferably, the method further includes a step of determining whether or not the inspection region is a region to be extracted, and the step of generating the changed inspection region image is performed when the inspection region is determined not to be the region to be extracted. A change inspection area image in which the area is enlarged is generated.

  Preferably, the method further includes a step of determining whether or not the inspection region is a region to be extracted, and the step of calculating the feature amount of the inspection region image includes a step of calculating a centroid position of the inspection region image. In the step of inputting an image, when it is determined that the inspection region is a region to be extracted, a step of newly inputting a new teaching image having an inspection region and generating a new defect adding teaching image is a new step. A new defect addition teaching image in which the changed inspection area image is arranged at a position corresponding to the center of gravity position of the inspection area image in the teaching image is generated.

  Preferably, the method further includes a step of determining whether or not the inspection region is a region to be extracted, and the step of calculating the feature amount of the inspection region image includes a step of calculating a centroid position of the inspection region image. In the step of inputting an image, when it is determined that the inspection area is an area to be extracted, a new teaching image having an inspection area is newly input, and the inspection area of the new teaching image is an area to be extracted. A step of determining whether or not there is a step of generating a new defect addition teaching image, when it is determined that the inspection region of the new teaching image is not a region to be extracted, A new defect addition teaching image in which the changed inspection area image is arranged at a position corresponding to the center of gravity position of the inspection area image is generated.

  Preferably, in the step of generating the change inspection area image, when it is determined that the inspection area is an area to be extracted, a change inspection area image in which the area of the inspection area image is reduced is generated.

  Preferably, in the step of generating the change inspection region image, when it is determined that the inspection region is not a region to be extracted, a change inspection region image in which the area of the inspection region image is enlarged is generated.

  Preferably, the step of setting the first and second inspection surrounding regions around the inspection region image and the change inspection region image, the first density value of the inspection region image, and the region in the first inspection surrounding region, respectively. The second density value of the first background image in the background area other than the examination area image, and the third density value of the second background image in the background area other than the changed examination area image in the second examination surrounding area. And calculating a fourth density value of the change inspection region image based on the above.

  Preferably, the first density value is an average density value of all pixels of the inspection area image, the second density value is an average density value of all pixels of the first background image, and the third density value is the second density value. The average density value of all the pixels of the background image, the step of making the inspection area image the same size and shape as the change inspection area image, and the step of calculating the fourth density value of the change inspection area image are the change inspection area image Setting the density values of all the pixels to the density values of all the pixels of the inspection area image having the same size and shape as the change inspection area image, and the density values of all the pixels of the set change inspection area image If the absolute value of the difference between the difference between the average value and the third density value and the difference between the first density value and the second density value is not less than a predetermined value, the absolute value is less than the predetermined value. The absolute value is added to or subtracted from the density value of all pixels in the set change inspection area image. Further comprising the step of.

  According to another aspect of the present invention, there is provided a teaching image generating apparatus for generating a teaching image having an inspection region for causing an algorithm that learns a target process in a self-developed manner in the teaching process. An image input unit for inputting a teaching image, an inspection region specifying unit for specifying an inspection region, a feature amount calculating unit for calculating a feature amount of an inspection region image in the specified inspection region, and a teaching image A position calculation unit that calculates a position where the image is to be arranged; a change inspection region image generation unit that generates a change inspection region image in which the feature amount of the inspection region image is changed; An image generation unit that generates a new defect addition teaching image in which the image is arranged.

  Preferably, the apparatus further includes an input unit that receives an instruction from the user, and the inspection area specifying unit specifies the inspection area based on the received user instruction.

Preferably, the position where the image is arranged is a position that does not overlap the inspection region image.
Preferably, the feature amount calculation unit calculates a centroid position of the inspection region image, and the position calculation unit calculates a position where the image is to be arranged by searching from the vicinity around the calculated centroid position of the inspection region image. .

  Preferably, an inspection region determination unit that determines whether or not the inspection region is a region to be extracted is further included, and the change inspection region image generation unit performs inspection when it is determined that the inspection region is a region to be extracted. A change inspection area image in which the area of the area image is reduced is generated.

  Preferably, an inspection area determination unit that determines whether the inspection area is an area to be extracted is further included, and the change inspection area image generation unit determines that the inspection area is not an area to be extracted. A change inspection area image in which the area of the image is enlarged is generated.

  Preferably, the image processing apparatus further includes an inspection region determination unit that determines whether or not the inspection region is a region to be extracted, the feature amount calculation unit calculates a center of gravity position of the inspection region image, and the image input unit includes the inspection region When it is determined that the region is to be extracted, a new teaching image having an inspection region is newly input, and the image generation unit is set to a position corresponding to the center of gravity position of the inspection region image in the new teaching image. Then, a new defect addition teaching image in which the change inspection area image is arranged is generated.

  Preferably, the image processing apparatus further includes a first inspection region determination unit that determines whether or not the inspection region is a region to be extracted, the feature amount calculation unit calculates a center of gravity position of the inspection region image, and the image input unit When it is determined that the area is an area to be extracted, a new teaching image having an inspection area is newly input, and it is determined whether the inspection area of the new teaching image is an area to be extracted. 2 an inspection area determination unit, and when the image generation unit determines that the inspection area of the new teaching image is not an area to be extracted, the position corresponding to the barycentric position of the inspection area image in the new teaching image Then, a new defect addition teaching image in which the change inspection area image is arranged is generated.

  Preferably, an area setting unit for setting the first and second examination surrounding areas around the examination area image and the change examination area image, the first density value of the examination area image, and the first examination surrounding area, The second density value of the first background image in the background area other than the inspection area image and the third density of the second background image in the background area other than the change inspection area image in the second inspection surrounding area. A density value calculation unit that calculates a fourth density value of the change inspection region image based on the density value is further provided.

  According to still another aspect of the present invention, teaching for causing a computer to execute processing for generating a teaching image having an inspection area for causing an algorithm that learns a target processing in a self-developed manner in the teaching process. The image generation program includes a step of inputting a teaching image having an inspection area, a step of specifying the inspection area, a step of calculating a feature amount of the inspection area image in the specified inspection area, Calculating a position for arranging the image in step, generating a changed inspection area image in which the feature amount of the inspection area image is changed, and a new defect in which the changed inspection area image is arranged at the position where the calculated image is arranged Generating an additional teaching image.

  According to still another aspect of the present invention, the recording medium is a medium on which a teaching image generation program is recorded.

  According to still another aspect of the present invention, a new defect addition teaching image having one or more inspection areas generated by a teaching image generating method using a plurality of types of image processing filters prepared in advance. An image processing algorithm generation method for generating one or more final image processing algorithms for generating an image corresponding to a target image prepared in advance includes a step of generating a plurality of image processing algorithms until a predetermined end condition is satisfied. The image processing algorithm generated in a generation that satisfies a predetermined end condition, including a step of repeating while changing the generation, is one or more final image processing algorithms, and each of the plurality of image processing algorithms includes a plurality of types of images. Algorithms that use multiple selected image processing filters selected in a random order without interfering with processing filters The process of generating a plurality of image processing algorithms of N (natural number) +1 generation is performed after a plurality of processes of N generations, each of which has been subjected to processing of a plurality of image processing algorithms of N generations on a new defect addition teaching image. A step of generating an image, a step of calculating an evaluation value serving as a reference for evaluation based on a target image and a corresponding processed image of N generations for each of a plurality of N generation image processing algorithms; Of the plurality of N generation image processing algorithms, a step of selecting one or more N generation image processing algorithms for which an evaluation value equal to or greater than a predetermined value is selected, and one or more of the selected one or more N generation image processing algorithms Generating an image processing algorithm that has been changed so that the evaluation value becomes high as an image processing algorithm of the N + 1 generation.

  Preferably, the predetermined end condition is a condition in which an evaluation value of any one or more of the N generations of selected image processing algorithms is equal to or greater than a predetermined value.

  Preferably, among the plurality of image processing algorithms, one or more image processing algorithms other than the N generation selected one or more image processing algorithms are generated as an N + 1 generation one or more image processing algorithms using an evolutionary method. The method further includes the step of:

Preferably, the evolutionary approach is a genetic algorithm.
Preferably, the number of new defect addition teaching images generated by the teaching image generation method is larger than the number of inspection areas in the new defect addition teaching image.

  According to still another aspect of the present invention, a new defect addition teaching image having one or more inspection areas generated by a teaching image generating method using a plurality of types of image processing filters prepared in advance. An image processing algorithm generation device that generates one or more final image processing algorithms that generate an image corresponding to a target image prepared in advance includes an image processing algorithm generation unit for generating a plurality of image processing algorithms. The image processing algorithm generation unit repeats the process of generating a plurality of image processing algorithms while changing the generation until a predetermined end condition is satisfied, and the image processing algorithm generated in the generation that satisfies the predetermined end condition is 1 The final image processing algorithm described above, and each of the plurality of image processing algorithms includes a plurality of types of image processing algorithms. A process for generating a plurality of image processing algorithms of N (natural number) +1 generation is used for new defect addition teaching. An image data conversion unit that generates a plurality of N generation processed images, each of which has been subjected to processing of a plurality of N generation image processing algorithms, and a target image for each of the N generation image processing algorithms And an image processing algorithm performance evaluation unit that calculates an evaluation value serving as a reference for evaluation based on the corresponding N generation processed images, and the image processing algorithm performance evaluation unit includes a plurality of N generations of image processing Among the algorithms, one or more N generations of image processing algorithms for which an evaluation value equal to or greater than a predetermined value is selected, and the image processing algorithm generation unit, Each of the 4-option by one or more N plurality of image processing algorithms generations were, the image processing algorithm were varied as evaluation value increases generated as image processing algorithms N + 1 generation.

  Preferably, the predetermined end condition is a condition in which an evaluation value of any one or more of the N generations of selected image processing algorithms is equal to or greater than a predetermined value.

  Preferably, among the plurality of image processing algorithms, one or more image processing algorithms other than the N generation selected one or more image processing algorithms are generated as an N + 1 generation one or more image processing algorithms using an evolutionary method. And an evolutionary image processing algorithm generating unit.

  Preferably, the number of new defect addition teaching images generated by the teaching image generation method is larger than the number of inspection areas in the new defect addition teaching image.

  According to still another aspect of the present invention, a new defect addition teaching image having one or more inspection areas generated by a teaching image generation program using a plurality of types of image processing filters prepared in advance. An image processing algorithm generation program for causing a computer to execute a process of generating one or more final image processing algorithms for generating an image corresponding to a target image prepared in advance includes a step of generating a plurality of image processing algorithms; The step of generating a plurality of image processing algorithms is repeated while changing the generation until a predetermined end condition is satisfied, and the image processing algorithm generated in the generation that satisfies the predetermined end condition is: One or more final image processing algorithms, each of a plurality of image processing algorithms Is an algorithm that uses a plurality of selected image processing filters selected from a plurality of types of image processing filters without interfering in a random order, and generates a plurality of N (natural number) +1 generation image processing algorithms. Generating a plurality of N generations of post-processing images obtained by performing processing of a plurality of N generations of image processing algorithms on the new defect addition teaching image, and for each of the N generations of image processing algorithms A step of calculating an evaluation value serving as a reference for evaluation based on the target image and the corresponding N generation processed images, and an evaluation value equal to or greater than a predetermined value among a plurality of N generation image processing algorithms is calculated. Selecting one or more selected N generation image processing algorithms, and selecting each of the selected one or more N generation image processing algorithms. Value changes image processing algorithms were so increases and generating an image processing algorithms N + 1 generation.

  According to still another aspect of the present invention, the recording medium is a medium on which an image processing algorithm generation program is recorded.

  According to still another aspect of the present invention, the image inspection method includes a step of determining pass / fail of the inspection target image using the image processing algorithm generated by the image processing algorithm generation method.

  According to still another aspect of the present invention, the image inspection apparatus includes an image quality determination unit that determines the quality of the inspection target image using the image processing algorithm generated by the image processing algorithm generation method.

  According to still another aspect of the present invention, the image inspection program causes the computer to execute a step of determining the quality of the inspection target image using the image processing algorithm generated by the image processing algorithm generation program.

  According to still another aspect of the present invention, the recording medium is a medium on which an image inspection program is recorded.

  In the teaching image generation method according to the present invention, in the teaching image having the inspection region, the feature amount of the inspection region image in the designated inspection region is calculated, and the inspection region image is arranged at the position where the calculated image is arranged. A new defect addition teaching image in which a changed inspection area image in which the feature amount is changed is generated.

  Therefore, since the teaching image generated by the present invention for causing the algorithm to learn the target process in a self-developed manner in the teaching process has a plurality of types of inspection region images, a plurality of types of inspection region images are obtained. There is an effect that the algorithm can be processed at a higher speed than using a teaching image having each of the above.

  The teaching image generation apparatus according to the present invention calculates a feature amount of an inspection region image in a specified inspection region in a teaching image having an inspection region, and places the calculated image at a position where the calculated image is arranged. A new defect addition teaching image in which a changed inspection area image in which the feature amount is changed is generated.

  Therefore, since the teaching image generated by the present invention for causing the algorithm to learn the target process in a self-developed manner in the teaching process has a plurality of types of inspection region images, a plurality of types of inspection region images are obtained. There is an effect that the algorithm can be processed at a higher speed than using a teaching image having each of the above.

  The teaching image generation program according to the present invention calculates the feature amount of the inspection region image in the designated inspection region in the teaching image having the inspection region, and at the position where the calculated image is arranged, the inspection region image A new defect addition teaching image in which a changed inspection area image in which the feature amount is changed is generated.

  Therefore, since the teaching image generated by the present invention for causing the algorithm to learn the target process in a self-developed manner in the teaching process has a plurality of types of inspection region images, a plurality of types of inspection region images are obtained. There is an effect that the algorithm can be processed at a higher speed than using a teaching image having each of the above.

  A recording medium according to the present invention records a teaching image generation program. The teaching image generation program calculates the feature amount of the inspection region image in the designated inspection region in the teaching image having the inspection region, and sets the feature amount of the inspection region image at the position where the calculated image is arranged. A new defect addition teaching image in which the changed inspection area image that has been changed is arranged is generated.

  Therefore, since the teaching image generated by the present invention for causing the algorithm to learn the target process in a self-developed manner in the teaching process has a plurality of types of inspection region images, a plurality of types of inspection region images are obtained. There is an effect that the algorithm can be processed at a higher speed than using a teaching image having each of the above.

  The image processing algorithm generation method according to the present invention generates an image processing algorithm using an inspection area image with high processing efficiency generated by the teaching image generation method.

  Therefore, an image processing algorithm having an evaluation value equal to or greater than a predetermined value can be generated at high speed.

  The image processing algorithm generation apparatus according to the present invention generates an image processing algorithm using an inspection area image with high processing efficiency generated by the teaching image generation method.

  Therefore, an image processing algorithm having an evaluation value equal to or greater than a predetermined value can be generated at high speed.

  The image processing algorithm generation program according to the present invention generates an image processing algorithm using an inspection area image with high processing efficiency generated by the teaching image generation program.

  Therefore, an image processing algorithm having an evaluation value equal to or greater than a predetermined value can be generated at high speed.

  The recording medium according to the present invention records an image processing algorithm generation program. The image processing algorithm generation program generates an image processing algorithm using an inspection region image with high processing efficiency generated by the teaching image generation program.

  Therefore, an image processing algorithm having an evaluation value equal to or greater than a predetermined value can be generated at high speed.

  The image inspection method according to the present invention determines the quality of the inspection target image using the image processing algorithm generated by the image processing algorithm generation method.

Therefore, the quality of the inspection target image can be determined appropriately.
The image inspection apparatus according to the present invention determines the quality of the inspection target image using the image processing algorithm generated by the image processing algorithm generation method.

Therefore, the quality of the inspection target image can be determined appropriately.
The image inspection program according to the present invention determines the quality of the inspection target image using the image processing algorithm generated by the image processing algorithm generation program.

Therefore, the quality of the inspection target image can be determined appropriately.
The recording medium according to the present invention records an image inspection program. The image inspection program determines the quality of the inspection target image using the image processing algorithm generated by the image processing algorithm generation program.

  Therefore, the quality of the inspection target image can be determined appropriately.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the following description, the same parts are denoted by the same reference numerals. Their names and functions are also the same. Therefore, detailed description thereof will not be repeated.

<First Embodiment>
FIG. 1 is a block diagram showing an internal configuration of a teaching image generation apparatus 500 according to the present embodiment. The teaching image generation apparatus 500 is a computer such as a PC (Personal Computer) or a workstation. Note that FIG. 1 also shows a recording medium 555 and an image data input unit 600 for explanation. The recording medium 555 stores a teaching image generation program 180 described later. That is, the teaching image generation program 180 is recorded on a medium or the like and distributed as a program product.

  Although details will be described later, the image data input unit 600 has a function of transmitting image data to the teaching image generation apparatus 500.

  With reference to FIG. 1, display unit 530, mouse 542, and keyboard 544 are connected to teaching image generation apparatus 500.

  The display unit 530 has a function of displaying various types of information as characters, images, and the like to the user. Display unit 530 displays an image based on the image data output from teaching image generation apparatus 500. The display unit 530 is a liquid crystal display (LCD (Liquid Crystal Display)), a CRT (Cathode Ray Tube), an FED (Field Emission Display), a PDP (Plasma Display Panel), an organic EL display (Organic Electroluminescence Display), a dot matrix, or the like. Any other display device of an image display method may be used.

  The mouse 542 is an interface for the user to operate the teaching image generation apparatus 500. The keyboard 544 is an interface for the user to operate the teaching image generation apparatus 500. The display unit 530 and the keyboard 544 may be provided inside the teaching image generation apparatus 500. Further, the mouse 542 may be configured to be provided inside the teaching image generation apparatus 500 as, for example, a touch pad. In addition, a pen tablet may be connected to the teaching image generation apparatus 500 as an input device.

  The teaching image generation apparatus 500 includes a control unit 510, a data temporary storage unit 522, a storage unit 520, a communication unit 560, a communication unit 562, a VDP (Video Display Processor) 532, and a CGROM (Character Graphic Read Only). Memory 534, VRAM (Video Random Access Memory) 536, an input unit 540, and a recording medium access unit 550.

  The CGROM 534 stores image data such as font data and graphic data for the VDP 532 to generate an image displayed on the display unit 530.

  The storage unit 520 stores a teaching image generation program 180 for causing the control unit 510 to perform processing to be described later, a teaching image database 182 for causing the image processing algorithm to learn, and various other programs and data. . The storage unit 520 is accessed by the control unit 510. The teaching image database 182 is a database composed of data of a plurality of types of teaching images that are different from each other. Details of the teaching image will be described later.

  The storage unit 520 is a hard disk capable of storing a large amount of data. Note that the storage unit 520 is not limited to a hard disk, and may be any medium (for example, a flash memory) that can hold data without being supplied with power.

  That is, the storage unit 520 uses an EEPROM (Erasable Programmable Read Only Memory) capable of erasing and writing the memory any number of times, an EEPROM (Electronically Erasable and Programmable Read Only Memory) capable of being electrically rewritten, and ultraviolet rays. Any of a UV-EPROM (Ultra-Violet Erasable Programmable Read Only Memory) capable of erasing and rewriting stored contents any number of times and a circuit having a configuration capable of storing and storing data in a nonvolatile manner may be used.

  Control unit 510 has a function of performing various types of processing, arithmetic processing, and the like for each device inside teaching image generation apparatus 500 in accordance with teaching image generation program 180 stored in storage unit 520. The control unit 510 includes a microprocessor, an FPGA (Field Programmable Gate Array), which is an LSI (Large Scale Integration) that can be programmed, and an ASIC (Application, which is an integrated circuit designed and manufactured for a specific application. Specific Integrated Circuit) or any other circuit having an arithmetic function may be used.

  The control unit 510 also instructs the VDP 532 to generate an image and display the image on the display unit 530 in accordance with the teaching image generation program 180 stored in the storage unit 520 (hereinafter, “drawing instruction”). Also called).

  The VDP 532 is connected to the display unit 530. The VDP 532 reads necessary image data from the CGROM 534 in accordance with a drawing instruction from the control unit 510 and generates an image using the VRAM 536. The VDP 532 reads out the image data stored in the VRAM 536 and causes the display unit 530 to display an image based on the image data.

The VRAM 536 has a function of temporarily storing an image generated by the VDP 532.
The data temporary storage unit 522 is accessed as data by the control unit 510 and used as a work memory for temporarily storing data.

  The data temporary storage unit 522 is a high-speed RAM (Random Access Memory), SRAM (Static Random Access Memory), DRAM (Dynamic Random Access Memory), SDRAM (Synchronous DRAM), or double data rate mode capable of temporarily storing data. DDR-SDRAM (Double Data Rate SDRAM), which is an SDRAM with excellent data transfer function, RDRAM (Rambus Dynamic Random Access Memory), DRAM that adopts high-speed interface technology developed by Rambus, Direct-RDRAM (Direct Rambus Dynamic) Random Access Memory) or any other circuit having a configuration capable of storing and storing data in a volatile manner.

  A mouse 542 and a keyboard 544 are connected to the input unit 540. The user gives an instruction to the teaching image generation apparatus 500 using the mouse 542 or the keyboard 544. An input instruction from the mouse 542 or the keyboard 544 is transmitted to the control unit 510 via the input unit 540. Control unit 510 performs predetermined processing based on an input instruction from input unit 540.

  The recording medium access unit 550 has a function of reading the teaching image generation program 180 from the recording medium 555 on which the teaching image generation program 180 is recorded. The teaching image generation program 180 stored in the recording medium 555 is read from the recording medium access unit 550 and stored in the storage unit 520 by the operation (installation process) of the control unit 510.

  The installation processing program is stored in advance in the storage unit 520, and the installation processing is performed by the control unit 510 based on the installation processing program.

  Note that the teaching image generation program 180 may not be installed in the storage unit 520. In this case, the control unit 510 reads the teaching image generation program 180 stored in the recording medium 555 via the recording medium access unit 550, and performs predetermined processing based on the teaching image generation program 180.

  The recording medium 555 includes a DVD-ROM (Digital Versatile Disk Read Only Memory), a CD-ROM (Compact Disk Read Only Memory), an MO (Magneto Optical Disk), a floppy (registered trademark) disk, a CF (Compact Flash) card, and an SM. (Smart Media (registered trademark)), MMC (Multi Media Card), SD (Secure Digital) memory card, Memory Stick (registered trademark), xD picture card and USB memory, magnetic tape, and other non-volatile memory Good.

  Communication unit 560 exchanges data with control unit 510. Communication unit 560 exchanges data with image data input unit 600 in a wired or wireless manner.

  Control unit 510 transmits control data for controlling image data input unit 600 via communication unit 560. Control unit 510 receives data corresponding to the transmitted control data from image data input unit 600 via communication unit 560.

  The communication unit 560 may be any of USB (Universal Serial Bus) 1.1, USB 2.0, and other communication interfaces for performing serial transfer. The communication unit 560 may be a communication interface using Ethernet (registered trademark). The communication unit 560 may be any of a Centronics specification, IEEE 1284 (Institute of Electrical and Electronic Engineers 1284), and other communication interfaces that perform parallel transfer. The communication unit 560 may be any one of communication interfaces using IEEE 1394 or other SCSI standards.

  The communication unit 560 may be any of wireless LAN standards IEEE802.11a, IEEE802.11b, IEEE802.11g, and other communication interfaces for performing data communication using a wireless technology.

  Communication unit 562 exchanges data with control unit 510. Since communication unit 562 is similar to communication unit 560, detailed description will not be repeated.

  Communication unit 610 exchanges data with communication unit 562 and network 620. The network 150 is an external network such as the Internet. The communication unit 610 is a communication interface (for example, a router) using Ethernet (registered trademark). Communication unit 610 and network 150 exchange data wirelessly or by wire. The wire is, for example, a telephone line or an optical fiber cable.

  Therefore, teaching image generation apparatus 500 in the present embodiment can also download a program from network 150 via communication unit 610 and communication unit 562 and store the program in storage unit 520. In this case, the downloaded program is the teaching image generation program 180.

  Control unit 510 performs predetermined processing in accordance with a program downloaded from network 150 (teaching image generation program 180). The download program is stored in advance in the storage unit 520, and the download process is performed by the control unit 510 based on the download program.

A teaching image is generated by the teaching image generating apparatus 500 having the above configuration.
FIG. 2 is a block diagram showing the internal configuration of the image data input unit 600. Note that FIG. 2 also shows a communication unit 560 in the teaching image generation apparatus 500 for explanation.

  Referring to FIG. 2, image data input unit 600 has a function of irradiating light to inspection object 650 to be inspected installed on stage 640 and converting reflected light of inspection object 650 into data. The inspection object 650 may be a liquid crystal panel, a semiconductor, an electronic component, plastic, metal, wood, paper, cloth, or the like.

  The image data input unit 600 includes a control unit 610, a light source 620, and a camera 630. Control unit 610 exchanges data with communication unit 560 in teaching image generation apparatus 500. Control unit 610 has the same configuration and function as control unit 510 described above. Control unit 610 controls light source 620 and camera 630 based on control data received from control unit 510 via communication unit 560.

The light source 620 applies light to the inspection object 650 under the control of the control unit 610.
The camera 630 has a function of imaging reflected light from the inspection object 650 under the control of the control unit 610. The camera 630 may be either a (Charge Coupled Devices) CCD system, a (Complementary Metal Oxide Semiconductor) CMOS system, or another system camera.

  In addition, the camera 630 transmits the captured data to the control unit 610 as image data. Control unit 610 transmits the received image data to control unit 510 via communication unit 560. The control unit 510 stores the received image data in the storage unit 520 as a teaching original image to be described later. Control unit 510 performs processing for displaying the received image data on display unit 530. The inspector (user) can determine whether or not the inspection object 650 is a defective product by referring to an image (hereinafter, also referred to as an inspection image) displayed on the display unit 530. Note that the inspector (user) may directly check the inspection object 650 and determine whether or not it is a defective product.

  Here, the terms used below are defined. First, a genuine defect that causes an inspector (user) to determine that the inspection object 650 is defective is defined as a true defect. On the other hand, although it is a defect extracted in the inspection process, the inspector (user) does not determine that the inspection object 650 is a defective product, and a defect that can be determined that the inspection target is a non-defective product is a pseudo defect. And

  Next, among the images obtained by imaging the inspection object 650, an image in which no true defect exists in the image regardless of the presence or absence of a pseudo defect is defined as a non-defective image. At least one existing image is defined as a defect image. The non-defective image and the defect image can be teaching images.

  Assume that the size of an image used in the following embodiments is, for example, horizontal 640 pixels and vertical 480 pixels. Note that the size of an image used in the following embodiments is not limited to 640 pixels wide and 480 pixels high, and may be any size (eg, 320 pixels wide and 240 pixels high).

  Further, the density values of the pixels of the image used in the following embodiments are values in the range of “0” to “255”. If the density value is “0”, the corresponding pixel is assumed to be black. If the density value is “255”, the corresponding pixel is white. Note that the density values of the pixels of the image used in the following embodiments are not limited to values in the range of “0” to “255”, and may be values in an arbitrary range.

FIG. 3 is a diagram showing a teaching image 600 as an example.
Referring to FIG. 3, teaching image 600 includes a defective area 660. The defect area 660 includes a defect itself area 620. Here, the defect itself region is a region indicating the defect itself. The defect itself region is classified into a true defect itself region which is a region showing a true defect and a pseudo defect itself region which is a region showing a pseudo defect. Hereinafter, the true defect itself region is also simply referred to as a true defect. Further, the pseudo defect itself region is also simply referred to as a pseudo defect. Also, true defects and pseudo defects are collectively referred to as defects.

  An area other than the defect itself area 620 in the teaching image 600 is a background area. That is, the teaching image 600 includes a background area and a defect area. Further, a region other than the defect itself region 620 in the defect region 660 is a defect background region 662. In the present embodiment, the defect area is a rectangular area. The shape of the defect area is not limited to a rectangle, and may be, for example, a circle or an ellipse.

  The defect area 660 is a rectangular area having a size that is twice the diameter of the defect itself 620 in the horizontal and vertical directions in the image, with the center of gravity of the defect area 620 as the center of the rectangular area. However, when the defect area 660 exists at the edge of the image, the edge of the image is given priority, and the rectangular area that is the defect area 660 is smaller than twice the size of the ferret diameter.

  The teaching images are classified into teaching original images and teaching target images. Regardless of the presence or absence of a true defect, an image obtained by imaging the inspection object 650 is used as a teaching original image, and an image provided with information on the true defect itself area in the teaching original image is referred to as a teaching target image. To do. The teaching original image and the teaching target image corresponding to the teaching original image are stored in the storage unit 520 in association with each other.

  Here, the teaching original image is an image subjected to image processing by an image processing algorithm. The teaching target image is an image for which it is desired that an image obtained by performing image processing with an image processing algorithm on the teaching original image matches or approximates.

  The target image for teaching in the present embodiment is a binary image in which the density value of each pixel is “0” or “255”, that is, black or white. The teaching target image is an image in which only defective pixels in the teaching target image are white and other pixels are black.

  FIG. 4 is a diagram showing a teaching original image 700 and a teaching target image 700A as an example.

  FIG. 4A shows a teaching original image 700 as an example. The teaching original image 700 includes a true defect 720 and a pseudo defect 710.

  FIG. 4B shows a teaching target image 700A as an example. The teaching target image 700A includes a true defect corresponding image 720A corresponding to the true defect 720.

  Next, processing for generating an image in which a defect is further added to the teaching original image using the defect in the teaching original image by the teaching image generating apparatus 500 will be described. In the following, the defect to be added to the teaching original image is also referred to as an adding source teaching defect. A defect newly added based on the additional source teaching defect is also referred to as a new additional teaching defect. Further, an image obtained by adding a new additional teaching defect to the teaching original image using the additional original teaching defect in the teaching original image is also referred to as a new defect adding teaching image. The process of generating a new defect addition teaching image is also referred to as a new defect addition teaching image generation process.

Next, a new defect addition teaching image generation process will be described.
FIG. 5 is a flowchart of a new defect addition teaching image generation process.

  Referring to FIG. 5, in step S <b> 100, control unit 510 reads teaching original image data from storage unit 520. Control unit 510 may input teaching original image data, which is an image of inspection object 650, from image data input unit 600. Assume that the read teaching original image is, for example, the teaching original image 700 of FIG.

  Hereinafter, a defect (for example, a defect corresponding to each of the true defect 720 and the pseudo defect 710) included in the teaching original image 700 read in step S100 is referred to as an initial teaching defect. The initial teaching defect that is a true defect is referred to as an initial teaching true defect. The initial teaching defect that is a pseudo defect is referred to as an initial teaching pseudo defect.

  In this embodiment, a case where one teaching original image is input will be described. However, the present embodiment is not limited to this, and a plurality of teaching original images may be input as necessary. In this case, the processing described below is repeated for the number of teaching original images.

When the process of step S100 ends, the process proceeds to step S102.
In step S <b> 102, control unit 510 sends a drawing instruction for displaying teaching original image 700 to display unit 530 to VDP 532. The VDP 532 displays the teaching original image 700 on the display unit 530 based on the drawing instruction. Thereafter, the process proceeds to step S104.

  In step S104, additional teaching defect selection processing is performed. In the additional teaching defect selection process, control unit 510 selects true defect 720 and pseudo defect 710 in teaching original image 700 read in step S100. Specifically, the user refers to the teaching original image 700 displayed on the display unit 530 and uses an input device such as the mouse 542, for example, each contour portion of the true defect 720 and the pseudo defect 710. For example, a selection process for selecting two defects by surrounding a defect by drawing a line is performed. Based on the selection process, control unit 510 causes storage unit 520 to store line data (hereinafter also referred to as line data) surrounding two defects.

  In the selection process, the control unit 510 may automatically create the line data based on the teaching original image 700. The true defect 720 and the pseudo defect 710 selected by the processing in step S104 are additional source teaching defects. All initial teaching defects are classified into initial teaching true defects corresponding to true defects and initial teaching pseudo defects corresponding to pseudo defects. Thereafter, the process proceeds to step S106.

  In step S106, an additional source teaching defect designation image generation process is performed. In the additional source teaching defect designation image generation process, the control unit 510 reads line data from the storage unit 520, and generates an additional source teaching defect designation image based on the line data. The control unit 510 causes the storage unit 520 to store the data of the additional source teaching defect designation image in association with the teaching original image. Note that the process of step S106 may not be performed.

  FIG. 6 shows a teaching original image 700 and an additional source teaching defect designation image 705A generated based on the teaching original image 700.

FIG. 6A shows a teaching original image 700.
FIG. 6B shows an additional source teaching defect designation image 705A. The additional source teaching defect designation image 705A is a binary image of white and black. The added source teaching defect designation image 705A includes a true defect corresponding image 720A corresponding to the true defect 720 and a pseudo defect corresponding image 710A corresponding to the pseudo defect 710A. The true defect corresponding image 720 </ b> A is based on a line drawn by the user on the contour portion of the true defect 720. The pseudo defect corresponding image 710 </ b> A is based on a line drawn by the user on the contour portion of the pseudo defect 710.

  Referring to FIG. 5 again, when the process of step S106 is completed, the process proceeds to step S108.

  In step S108, a defective area setting process is performed. In the defect area setting process, control unit 510 calculates the ferret diameter of each of true defect corresponding image 720A and pseudo defect corresponding image 710A. Then, control unit 510 sets a defect region for each of true defect 720 and pseudo defect 710 of teaching original image 700 based on the calculated ferret diameter.

FIG. 7 is a defect area setting image in which a defect area is set for the teaching original image 700.
Referring to FIG. 7, defect area 760 and defect area 750 are set in teaching original image 700. The defect area 760 is an area set for the true defect 720. The defect area 750 is an area set for the pseudo defect 710.

  Referring to FIG. 5 again, when the process of step S108 ends, the process proceeds to step S110.

  In step S110, a feature amount calculation process is performed. In the feature amount calculation process, control unit 510 calculates the feature amounts of defect area 760 and defect area 750. The feature amount includes the position of the center of gravity on the image, the area, the angle (rotation angle), the contrast, and the like of the defect in the defect region.

  For example, the feature amount of the defect area 760 is determined based on the true defect correspondence image 720A in the additional source teaching defect designation image 705A by the control unit 510 based on the center of gravity position, area, and angle (rotation angle) on the image. calculate.

  Note that the contrast in the present embodiment is the absolute value of the difference between the average density value of the defect itself in the defect area and the average density value of the defect background area.

  For example, when the control unit 510 calculates the contrast of the defect region 760, the average density value of all pixels of the defect 720 that is the defect itself region in the defect region 760 (hereinafter, also referred to as the defect itself region average density value). And the average density value of all the pixels of the image in the above-described defect background area in the defect area 760 (hereinafter, also referred to as the defect background area average density value).

  Then, the control unit 510 obtains the absolute value of the difference between the defect itself area average density value and the defect background area average density value, and uses the absolute value as the contrast of the defect area 760. Control unit 510 causes storage unit 520 to store the calculated feature amounts of defect area 760 and defect area 750 in association with the defect area. Thereafter, the process proceeds to step S112.

  In step S112, a newly added defect number determination process is performed. In the new additional defect number determination process, for each of the additional source teaching defects (true defect 720 and pseudo defect 710) selected in step S104, a defect in which which feature quantity of the additional source teaching defect is changed is taught. A process of determining the number of defects to be added is performed by determining whether to newly add to the original image 700 for use. The processing is performed by the control unit 510 reading out the feature amount change setting data for setting the number of newly added teaching defects stored in the storage unit 520.

  When the feature quantity to be changed is, for example, an area based on the feature quantity change setting data, and the area of the addition source teaching defect is reduced by 2% and 4%, respectively, a new one newly added to the teaching original image 700 There are two additional teaching defects.

  When the feature amount to be changed is, for example, contrast by the feature amount change setting data, and the contrast of the defect in the additional source teaching is reduced by 2% and 4%, respectively, a new newly added to the teaching original image 700 There are two additional teaching defects.

  That is, the number of newly added teaching defects is determined by designating the selected feature amount and each change amount.

The control unit 510 causes the storage unit 520 to store information on the determined number of defects.
In this process, the setting data is not used, and the user uses the input device such as the mouse 524 or the keyboard 544 in the GUI (Graphical User Interface) displayed on the display unit 530 to change the feature amount change setting data. It may be performed by inputting corresponding data. Thereafter, the process proceeds to step S114.

  In step S114, a newly added teaching defect list creation process is performed. In the new additional teaching defect list creation process, control unit 510 assigns additional source teaching defect IDs 1 and 2 to true defect 720 and pseudo defect 710 as additional source teaching defects, respectively. Then, based on the feature amount change setting data read in step S112 and the number of defects to be added, the control unit 510 assigns new additional teaching defect IDs in any order, for example, for all new additional teaching defects. 1 to 4 are assigned to create a new additional teaching defect list table. The newly added teaching defect list table is stored in the storage unit 520 by the control unit 510.

FIG. 8 is a diagram showing a new additional teaching defect list table T100 as an example.
Referring to FIG. 8, new additional teaching defect list table T100 shows an example in which the number of feature quantities to be simultaneously changed with respect to the additional source teaching defect is two. Note that the number of feature quantities to be changed may be three or more.

  The new additional teaching defect list table T100 is, for example, a list when the number of defects to be added is two for each of the two additional source teaching defects to which the additional source teaching defects ID1 and ID2 are respectively assigned. Show.

  The new additional teaching defect list table T100 is a list table in which each of a plurality of new additional teaching defect IDs, an additional source teaching defect ID, a feature amount to be changed, and a feature amount change amount are associated with each other. . Hereinafter, the data assigned to each new additional teaching defect ID is also referred to as new additional teaching defect data. Therefore, the newly added teaching defect list table T100 includes a plurality of newly added teaching defect data.

  Referring to FIG. 5 again, when the process of step S114 ends, the process proceeds to step S120.

  In step S120, "1" is set to the counter k for specifying data to be read from the newly added teaching defect list table T100. Thereafter, the process proceeds to step S122.

  In step S122, control unit 510 reads defect data for newly added teaching corresponding to counter k from storage unit 520. When the counter k is “1”, the control unit 510 reads from the storage unit 520 the new additional teaching defect data corresponding to the new additional teaching defect ID “1” in the new additional teaching defect list table T100.

  In this case, the new additional teaching defect based on the new additional teaching defect data is a defect in which only the area of the defect corresponding to the additional source teaching defect ID “1” is reduced by 5%.

  When the counter k is “4”, the control unit 510 stores new additional teaching defect data corresponding to the new additional teaching defect ID “4” in the new additional teaching defect list table T100 from the storage unit 520. read out.

  In this case, the new additional teaching defect based on the new additional teaching defect data is a defect in which the area of the defect corresponding to the additional source teaching defect ID “2” is reduced by 5% and the contrast is reduced by 5%. . When the process of step S122 ends, the process proceeds to step S130.

  In step S130, a new additional defect drawing position determination process for determining the drawing position of the new additional teaching defect is performed. The drawing position of the newly added teaching defect is expressed by the position of the center of gravity of the defect. In addition, the drawing position of the newly added teaching defect is determined based on whether the newly added teaching defect is an initial teaching defect in the teaching original image or a newly added teaching defect that has already been added. The position is determined so as not to overlap with the part. Further, the drawing position of the newly added teaching defect is determined at a position where the newly added teaching defect falls within the teaching original image.

  FIG. 9 is a flowchart of the newly added defect drawing position determination process. In the processing described below, for example, the teaching original image 700 of FIG. 4 is used.

  Referring to FIG. 9, in step S131, an initial setting process is performed. In the initial setting process, control unit 510 reads drawing position setting data set in advance by the user from storage unit 520. Then, initial setting is performed based on the read drawing position setting data.

  The drawing position setting data includes neighborhood selection data indicating whether or not the drawing position of the newly added teaching defect is selected from the vicinity of the adding source teaching defect. When the neighborhood selection data is set to “1”, for example, the drawing position of the newly added teaching defect is selected from the vicinity of the adding source teaching defect.

  Whether to use the density value dispersion value in the drawing position setting data when determining whether the drawing position where the additional source teaching defect is arranged is an appropriate position with respect to the background density value. Distributed value use permission data indicating whether or not is included. For example, when the dispersion value use permission data is set to “1”, the dispersion value of the density value is used in the determination.

When the process of step S131 ends, the process proceeds to step S132.
In step S132, control unit 510 determines whether or not to select the drawing position of the newly added teaching defect from the vicinity of the additional source teaching defect. Specifically, it is determined whether or not the read neighborhood selection data is “1”. If YES in step S132, the process proceeds to step S133. On the other hand, if NO at step S132, the process proceeds to step S133A.

  In step S133, control unit 510 selects the drawing position of the newly added teaching defect from the vicinity of the additional source teaching defect. Specifically, control unit 510 determines, for example, a position 4 pixels away from the barycentric position of the additional source teaching defect (for example, true defect 720) of teaching original image 700 as the drawing position. Note that each time the process of step S133 is performed, the controller 510 determines, for example, a position on the spiral that is 4 pixels away from the previously determined drawing position as the drawing position. Note that the method of determining the drawing position is not limited to the spiral. Further, the position 4 pixels away when determining the drawing position is an example, and may be a position 8 pixels away, for example. Thereafter, the process proceeds to step S133B.

  In step S133A, control unit 510 randomly selects the drawing position of the newly added teaching defect. The control unit 510 randomly determines the drawing position of the new additional teaching defect regardless of the position of the additional source teaching defect (for example, the true defect 720) of the teaching original image 700. Thereafter, the process proceeds to step S133B.

  Note that the process of selecting the drawing position of the newly added teaching defect is not limited to the processes of steps S133 and S133A. For example, the processing for selecting the drawing position of the newly added teaching defect may be performed in order from the upper left pixel of the image toward the right edge of the image.

  In step S133B, a new additional teaching defect creation process is performed. In the new additional teaching defect creation process, control unit 510 determines the size and shape of the new additional teaching defect based on the new additional teaching defect data read in step S122.

  In addition, when the size of the newly added teaching defect created by the control unit 510 based on the newly added teaching defect data is different from the size of the additional source teaching defect, the control unit 510 performs general image processing. The size and shape of the newly added teaching defect are determined by performing enlargement / reduction and aspect ratio change processing using the technology.

  When the new additional teaching defect created by the control unit 510 based on the new additional teaching defect data rotates with respect to the additional source teaching defect, the control unit 510 rotates using a general technique of image processing. By performing the conversion, the size and shape of the newly added teaching defect are determined. It should be noted that the density values of all the pixels of the newly added teaching defect are calculated by processing to be described later.

  FIG. 10 shows a new defect addition teaching image 705 in which a new additional teaching defect is arranged on the teaching original image 700.

  Referring to FIG. 10, new defect addition teaching image 705 is an example of an image in which new additional teaching defect 724 is arranged at the drawing position determined by the process of step S133 or S133A.

  Referring to FIG. 9 again, when the process of step S133B ends, the process proceeds to step S134.

  In step S134, control unit 510 determines whether or not the additional teaching defect arranged at the drawing position of the additional teaching defect determined in step S133 or step S133A overlaps with another defect.

  Specifically, the control unit 510 determines the coordinates of the plurality of pixels constituting the new additional teaching defect, which are determined based on the position of the additional original teaching defect, the size of the new additional teaching defect, and the like, It is determined whether or not the coordinates of each of a plurality of pixels constituting the addition source teaching defect or the newly added teaching defect that has already been added overlap.

  If YES in step S134, the process of step S132 is performed again. On the other hand, if NO at step S134, the process proceeds to step S135.

  In step S135, a defective area setting process is performed. In the defect area setting process, control unit 510 sets a defect area for newly added teaching defect 724. Specifically, control unit 510 calculates the number of horizontal pixels and the number of vertical pixels of the defect area based on the size information obtained from the drawing position, area, and ferret diameter information of additional teaching defect 724. Then, a defect area is set for the newly added teaching defect 724.

FIG. 11 is a defect area setting image in which a defect area is set for the teaching original image 705.
Referring to FIG. 11, defect area 762 is set in teaching original image 705. The defect area 762 is an area set for the newly added teaching defect 724.

  Referring again to FIG. 9, next, control unit 510 removes the pixel portion of new additional teaching defect 724 from the pixels in defective region 762, thereby newly adding new teaching defect corresponding to defective region 762. The pixels in the background area are calculated. Thereafter, the process proceeds to step S136.

  In step S136, an average density value comparison process is performed. In the average density value comparison process, control unit 510 calculates an average value of density values of all pixels in the newly added teaching defect background area (hereinafter, also referred to as a new additional teaching defect background area average density value). . Thereafter, the process proceeds to step S136A.

  In step S136A, control unit 510 determines the absolute value of the difference between the defect background area average density value in defect area 760 calculated in step S110 and the newly added teaching defect background area average density value calculated in step S136. Is less than a predetermined value (for example, 60).

  That is, in the process of steps S136 and S136A, it is determined whether or not the drawing position of the newly added teaching defect determined by the process of step S133 or S133A is an appropriate position.

  If YES in step S136A, the process proceeds to step S137. On the other hand, if NO at step S136A, the process at step S132 is performed again.

  In step S137, when the control unit 510 determines whether the drawing position of the new additional teaching defect determined by the process of step S133 or S133A is an appropriate position with respect to the density value of the background, Determine whether to use the variance value. Specifically, control unit 510 determines whether or not the distributed value use permission data is “1”. If YES at step S137, the process proceeds to step S137A. On the other hand, if “NO” in the step S137, the process of determining the new additional defect drawing position is ended, and the process returns to the new defect adding teaching image generation process again, and proceeds to the step S140 after the step S130.

  In step S137A, density value dispersion comparison processing is performed. In the density value variance comparison process, control unit 510 calculates a variance value of density values of all pixels in the defective background area in defective area 760 (hereinafter also referred to as an additional source teaching defect background area variance value).

  Next, control unit 510 calculates a variance value of density values of all pixels in the newly added teaching defect background region (hereinafter also referred to as a newly added teaching defect background region variance value). Thereafter, the process proceeds to step S137B.

  In step S137B, control unit 510 determines that the absolute value of the difference between the additional source teaching defect background area variance calculated in step S137A and the new additional teaching defect background area variance calculated in step S137A is predetermined. It is determined whether or not it is a value (for example, 60) or less.

  That is, in the processing of steps S137A and S137B, whether or not the drawing position of the newly added teaching defect determined by the processing of step S133 or S133A is a more appropriate position than the processing of steps S136 and S136A. Is determined.

  If “YES” in the step S137B, the process of the new additional defect drawing position determination process is ended, the process returns to the new defect additional teaching image generation process again, and proceeds to the step S140 subsequent to the step S130. On the other hand, if NO at step S137B, the process at step S132 is performed again.

  With the above-described new additional defect drawing position determination processing, the drawing position of the new additional teaching defect is determined.

  Referring to FIG. 5 again, after step S130, step S140 is performed.

  In step S140, a new additional teaching defect drawing process is performed. In the new additional teaching defect rendering process, the teaching original image 700 is rendered in the VRAM 536 for storing image data to be displayed on the display unit 530 by the processing of the control unit 510 and the VDP 532.

  Next, the density values of all the pixels of the newly added teaching defect 724 are calculated. Specifically, when the size and shape of the new additional teaching defect 724 and the additional source teaching defect 720 are the same, the control unit 510 determines the density values of all the pixels of the new additional teaching defect 724 as defects. The density value is set to 720.

  When the new additional teaching defect 724 is subjected to processing such as enlargement, reduction, and rotation, and the size and shape of the new additional teaching defect 724 and the additional source teaching defect 720 are different from each other, the control unit 510 performs the additional source teaching. The shape is changed by subjecting the defect 720 for enlargement, reduction, rotation, etc. to the same size and shape as the newly added teaching defect 724. Then, the density value of all pixels of the newly added teaching defect 724 is set to the density value of all pixels of the defect 720 whose shape has been changed.

  Next, the control unit 510 calculates an average value of all pixels of the new additional teaching defect 724 (hereinafter, also referred to as a new additional teaching defect average density value). Next, the control unit 510 calculates the absolute value of the difference between the new additional teaching defect average density value and the above-described new additional teaching defect background area average density value (hereinafter also referred to as new additional teaching defect contrast). Ask for.

  Next, if the absolute value of the difference between the newly added teaching defect contrast and the contrast of the defect area 760 is equal to or smaller than a predetermined value (for example, 0.4), the control unit 510 sets the newly added teaching. The density value of all pixels of the defect 724 for use is set as the final density value. The predetermined value may be “0”.

  In addition, if the absolute value of the difference between the newly added teaching defect contrast and the contrast of the defect area 760 is greater than a predetermined value, the control unit 510 causes the newly added teaching defect contrast to be equal to or less than the predetermined value. By adding or subtracting the absolute value (offset value) of the obtained difference to the density value of all pixels of the new additional teaching defect 724, the final density value of all pixels of the new additional teaching defect 724 is calculated. To do.

  Next, by the processing of the control unit 510 and the VDP 532, a shape is generated in step S133B at the drawing position of the new additional teaching defect determined by the new additional defect drawing position determination process on the VRAM 536, and the density value is obtained by the above process. The newly added teaching defect 724 for which is determined is drawn. Thereafter, the process proceeds to step S150.

  In step S150, control unit 510 determines whether the number of newly added source teaching defects determined in step S112 is equal to the number of newly added teaching defects drawn in step S140. If YES in step S150, the process proceeds to step S160. On the other hand, if NO at step S150, the process proceeds to step S152.

  In step S152, control unit 510 increments counter k by “1”. Thereafter, the process of step S122 is performed again.

  In step S160, control unit 510 causes storage unit 520 to store the image data (new defect addition teaching image data) stored in VRAM 536. Then, the new defect addition teaching image generation process ends.

  The new defect addition teaching image is generated by the new defect addition teaching image generation processing described above.

FIG. 12 is a functional block diagram of the teaching image generation apparatus 500 in the present embodiment.
Referring to FIG. 12, teaching image generation apparatus 500 operates as a teaching image generation management unit. An image data input unit 600 and a display unit 530 are connected to the teaching image generation apparatus 500.

Teaching image generation apparatus 500 includes a control unit 510.
The control unit 510 operates as an image input unit 511, a defect selection unit 512, an image processing calculation unit 513, a new additional teaching defect drawing unit 514, and a teaching image generation completion determination unit 516. The teaching image database 182 is stored in the storage unit 520. The teaching image database 182 includes the teaching original image input from the image data input unit 600.

The image input unit 511 has a function of performing the process of step S100 described above.
The defect selection unit 512 has a function of performing the process of step S104 described above.

  The image processing calculation unit 513 has a function of performing the processes of steps S108, S110, S112, S114, S120, S122, S130, and S152 described above.

  The newly added teaching defect drawing unit 514 has a function of performing the process of step S140 described above.

  The teaching image generation completion determination unit 516 has a function of performing the process of step S150 described above.

  The storage unit 520 operates as a teaching image storage unit that stores a new defect addition teaching image by the new defect addition teaching image generation processing.

  As described above, in the present embodiment, a new additional teaching defect in which the additional source teaching defect is determined from the teaching original image obtained by imaging the inspection object 650 and the feature amount of the additional source teaching defect is changed. Is added to the teaching original image, and a new defect adding teaching image is generated. Therefore, the new defect addition teaching image is for the new additional teaching that reflects the feature amount of the defect in the actual environment, as compared to the case where the new additional teaching defect is artificially drawn and added to the original teaching image. There is an effect that an image in which defects are arranged is obtained.

  In the present embodiment, a plurality of newly added teaching defects are added on one teaching original image. Therefore, since it is not necessary to create a plurality of images in which one new additional teaching defect is arranged and store it in the storage unit 520, the access time to the storage unit 520 can be shortened.

  In addition, by storing an image in which a plurality of new additional teaching defects are arranged in the storage unit 520, the storage area used by the storage unit 520 can be reduced, and as a result, cost reduction can be realized. The effect of becoming.

  Further, the drawing position of the newly added teaching defect is determined by searching around the position of the center of gravity of the additional source teaching defect in the original teaching image. Thus, when the drawing position of the new additional teaching defect is determined to be close to the position of the center of gravity of the additional original teaching defect, the defect background area of the new additional teaching defect is the defect background area of the additional original teaching defect. It will be similar to Therefore, there is an effect that it is possible to reflect the appearance of the defect depending on the location to be inspected.

  In addition, when a new additional teaching defect in which the density value of the additional source teaching defect is changed is generated, an offset of a difference between the average density value of the new additional teaching defect and the average density value of the additional source teaching defect is newly set. Addition or subtraction to additional teaching defect pixels.

  That is, the density distribution shape of the newly added teaching defect is the one that is translated along the density axis while maintaining the density distribution shape of the additional source teaching defect. Therefore, a new additional teaching defect having the same density distribution as that of the additional source teaching defect but having a different area and appearance position can be created.

  As a result, compared to the case where a new additional teaching defect is artificially added to the teaching original image, a new additional teaching defect reflecting the feature amount of the defect in the actual environment is added to the teaching original image. There is an effect that can be.

<Second Embodiment>
In the present embodiment, a process for generating a new defect addition teaching image more suitable for learning than in the first embodiment will be described. Hereinafter, the process of generating a new defect addition teaching image suitable for learning is also referred to as a new defect addition teaching image generation process A.

  FIG. 13 is a block diagram showing an internal configuration of teaching image generation apparatus 500A in the present embodiment.

  Referring to FIG. 13, teaching image generation apparatus 500 </ b> A has data access to recording medium 555 </ b> A instead of recording medium 555 as compared with teaching image generation apparatus 500, and teaching image is stored in storage unit 520. A difference is that a teaching image generation program 180A is recorded instead of the generation program 180 and a teaching image generation program 180A is recorded on the recording medium 555A. Since other configurations and functions are the same as those of teaching image generation apparatus 500, detailed description will not be repeated.

FIG. 14 is a flowchart of the image generation process A for new defect addition teaching.
Referring to FIG. 14, the new defect addition teaching image generation process A is different from the new defect addition teaching image generation process of FIG. 5 in that the process of step S103A is performed between the processes of step S102 and step S104. A point that is performed, a point that the process of step S112A is performed instead of step S112, a point that the process of step S114A is performed instead of step S114, a point that the process of step S120A is performed instead of step S120, The difference is that the process of step S122A is performed instead of step S122. Since the other processes are the same as the new defect addition teaching image generation process, detailed description will not be repeated.

  It is assumed that the teaching original image read in step S100 of the new defect addition teaching image generation process A is, for example, the teaching original image 700 of FIG.

  In the new defect addition teaching image generation process A, the process proceeds to step S103A after the process of step S102.

  In step S103A, a defect determination process is performed. In the defect determination processing, control unit 510 determines whether the defect in teaching original image 700 read in step S100 is a true defect or a pseudo defect.

  Specifically, the user refers to the teaching original image 700 displayed on the display unit 530 and sets the defect in the teaching original image 700 to either a true defect or a pseudo defect. The setting is performed by the user giving an instruction to the control unit 510 using an input device such as the mouse 542 or the keyboard 544.

  Control unit 510 performs the above-described defect determination process based on the instruction. In the present embodiment, it is assumed that the defect 720 of the teaching original image 700 is set as a true defect and the defect 710 is set as a pseudo defect. Hereinafter, data indicating the setting information of the defect is also referred to as defect setting data. The defect setting data is stored in the storage unit 520 by the control unit 510.

  In step S112A, a newly added defect count determination process A is performed. In the new additional defect number determination process A, a defect in which which feature amount of the additional source teaching defect is changed to how much is added to each of the additional source teaching defect (true defect 720 and pseudo defect 710) selected in step S104. Processing for determining the number of defects to be added is performed by determining whether to newly add to the teaching original image 700. The processing is performed by the control unit 510 reading out the feature amount change setting data for setting the number of newly added teaching defects stored in the storage unit 520.

  The feature amount change setting data in the present embodiment is data for specifying the number of new additional teaching defects in which the area of the additional source teaching defect is reduced when the additional source teaching defect is a true defect. .

  The feature amount change setting data is data for specifying the number of new additional teaching defects in which the area of the additional source teaching defect is increased when the additional source teaching defect is a pseudo defect.

  Here, the feature amount change setting data is data for reducing only the area of the additional source teaching defect that is a true defect, for example, by reducing the area by 10, 20%, respectively. is there. Therefore, the number of new additional teaching defects as true defects is two.

  In addition, the feature amount change setting data is data that increases the feature amount only for the area, for example, by 10% or 20% for the additional source teaching defect that is a pseudo defect. . Therefore, the number of new additional teaching defects as pseudo defects is two.

  Specifically, control unit 510 reads defect setting data from storage unit 520, recognizes the number of true defects and pseudo defects, and determines the number of new additional teaching defects based on the feature amount change setting data. . In the present embodiment, since the defect setting data indicates information on one true defect and one pseudo defect, the control unit 510 adds a new defect as a true defect based on the above-described feature amount change setting data. The number of teaching defects is determined to be two, and the number of newly added teaching defects as pseudo defects is determined to be two. The control unit 510 causes the storage unit 520 to store information on the determined number of defects. The defect setting data is not limited to data indicating information on one true defect and one pseudo defect. The defect setting data may be data indicating information on two or more true defects and two or more pseudo defects, for example. That is, the defect setting data changes according to the teaching original image read out in step S100.

  In step S114A, a newly added teaching defect list creation process A is performed. In the new additional teaching defect list creation process A, the control unit 510 assigns additional source teaching defects ID 1 and 2 from defects having a large area based on the defect area calculated in step S110. Since the true defect 720 is larger than the pseudo defect 710, the additional source teaching defects ID 1 and 2 are assigned to the true defect 720 and the pseudo defect 710, respectively.

  Next, based on the feature amount change setting data read in step S112A and the number of defects to be added, control unit 510 assigns new additional teaching defect IDs, for example, 1 to Allocate up to 4, and create a new additional teaching defect list table. The newly added teaching defect list table is stored in the storage unit 520 by the control unit 510.

FIG. 15 is a diagram showing a newly added teaching defect list table T200 as an example.
Referring to FIG. 15, new additional teaching defect list table T200 has two defects added to each of two additional source teaching defects assigned with additional source teaching defects ID1 and ID2, respectively. Here is a list of cases.

  Referring to FIG. 14 again, after the process of step S114A, the process proceeds to step S120A.

  In step S120A, control unit 510 sets counter k for specifying data to be read from newly added teaching defect list table T200 to “1”. Thereafter, the process proceeds to step S122A.

  In step S122A, control unit 510 reads out newly added teaching defect data corresponding to counter k from storage unit 520. When the counter k is, for example, “1”, the control unit 510 stores the newly added teaching defect data corresponding to the newly added teaching defect ID “1” in the newly added teaching defect list table T200. Read from.

  Then, similarly to the new defect addition teaching image generation processing of FIG. 5, the processing of steps S122A, S130, S140, and S152 is repeated until the condition of step S150 is satisfied, thereby generating a new defect addition teaching image. Is done.

  FIG. 16 is a diagram showing a new defect addition teaching image 700B generated in the present embodiment.

  Referring to FIG. 16, new defect addition teaching image 700 </ b> B has a new additional teaching defect as a true defect in which the area is reduced by 10, 20% with respect to true defect 720 as an addition source teaching defect, respectively. 722, 724 are arranged. Further, in the new defect addition teaching image 700B, new additional teaching defects 712 and 714 as pseudo defects, which are respectively increased by 10% and 20% with respect to the pseudo defect 710 as the additional source teaching defect, are arranged. Is done.

  In step S136A described above, control unit 510 causes new additional teaching defect so that the contrast of the defect area corresponding to the additional teaching defect is substantially equal to the contrast of the defect area corresponding to the additional source teaching defect. It is assumed that processing for changing the average density value is performed.

  The new additional teaching defect background area average density value of the new additional teaching defect 724 is higher than the average density value of the background area of the true defect 720 as the additional source teaching defect. Therefore, the average density value of the new additional teaching defect 724 is set higher than the average density value of the true defect 720 by the process of changing the new additional teaching defect average density value.

The new additional teaching defect 714 is set in the same manner as the new additional teaching defect 724.
As described above, in the present embodiment, when the additional source teaching defect is a true defect, a new additional teaching defect in which the area of the additional source teaching defect is reduced is generated. If the additional source teaching defect is a pseudo defect, a new additional teaching defect in which the area of the additional source teaching defect is increased is generated.

  Therefore, reflect the problem that the true defect becomes harder to detect as the defect becomes smaller, and the false defect should not be detected as the defect becomes larger, in the algorithm for learning the new defect addition teaching image. Can do. That is, it is possible to generate a new defect addition teaching image as a teaching image more suitable for learning.

  In addition, according to the present embodiment, since there is no need to add a new additional teaching defect, which is a large true defect or a small pseudo defect, which is not necessary as a teaching image, the time required for creating the teaching image is shortened. There is an effect that can be.

<Third Embodiment>
In the new defect addition teaching image generation processing or the new defect addition teaching image generation processing A of the first and second embodiments, the teaching original image is converted into the teaching original image using the additional source teaching defect of the teaching original image. A new defect addition teaching image was added. In the present embodiment, when the teaching original image is a defect image in which a true defect exists, processing for newly reading image data and adding a new defect addition teaching image to the read image will be described. . Hereinafter, this process is also referred to as a new defect addition teaching image generation process B.

  FIG. 17 is a block diagram showing an internal configuration of teaching image generation apparatus 500B in the present embodiment.

  Referring to FIG. 17, teaching image generation apparatus 500B performs data access to recording medium 555B instead of recording medium 555 as compared with teaching image generation apparatus 500 in FIG. The difference is that the teaching image generation program 180B is recorded instead of the teaching image generation program 180 and the teaching image generation program 180B is recorded on the recording medium 555B. Since other configurations and functions are the same as those of teaching image generation apparatus 500, detailed description will not be repeated.

FIG. 18 is a flowchart of the new defect addition teaching image generation process B.
Referring to FIG. 18, in step S300, as in step S100 of FIG. 5, control unit 510 reads teaching original image data. Assume that the read teaching original image is, for example, the teaching original image 700 of FIG. Thereafter, the process proceeds to step S302.

  In step S302, processing similar to that in step S102 is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S303.

  In step S303, defect determination processing is performed as in step S103A. That is, control unit 510 determines whether a defect in teaching original image 700 is a true defect or a pseudo defect. Since the specific process of the determination is the same as that in step S103A, detailed description will not be repeated.

  Next, when controller 510 determines that there is a true defect in teaching original image 700, controller 510 determines that the teaching original image is a defective image. On the other hand, if controller 510 determines that there is no true defect even if there is a pseudo defect in teaching original image 700, controller 510 determines that the teaching original image is a non-defective image. Control unit 510 causes storage unit 520 to store image determination data indicating information that the teaching original image is either a defect image or a non-defective image, and defect setting data. Thereafter, the process proceeds to step S310.

  In step S310, control unit 510 determines whether or not the teaching original image read in step S300 is a defective image by the process in step S303. If YES in step S310, the process proceeds to step S312. On the other hand, if NO at step S310, the process proceeds to step S314.

  In step S312, new defect addition image data is input. Specifically, control unit 510 newly reads teaching original image data in the same manner as step S100 in FIG. Hereinafter, in this process, the teaching original image based on the newly read teaching original image data is also referred to as a new defect adding image.

  FIG. 19 shows a new defect addition image and a new defect addition teaching image in the present embodiment.

  FIG. 19A is a diagram showing a new defect addition image 800 read out in the process of step S312. As an example, it is assumed that the new defect addition image 800 is a non-defective image having pseudo defects 810 but no true defects.

  FIG. 19B is a diagram showing a new defect addition teaching image 800A generated by the new defect addition teaching image generation processing B.

Referring to FIG. 18 again, after the process of step S312, the process proceeds to step S313.
In step S313, defect determination processing similar to that in step S303 is performed on the new defect addition image as the teaching original image. That is, control unit 510 determines whether the new defect addition image is a defect image or a non-defective image. Then, control unit 510 causes storage unit 520 to store image determination data indicating information that the new defect addition image is either a defect image or a non-defective image, and defect setting data. Thereafter, the process proceeds to step S314.

  In step S314, the same processing as in step S104 is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S316.

  In step S316, processing similar to that in step S106 is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S318.

  In step S318, the same processing as in step S108 is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S320.

  In step S320, the same processing as in step S110 is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S322.

  In step S322, processing similar to that in step S112A is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S324.

  In step S324, the same processing as in step S114A is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S330.

  In step S330, the same process as in step S120A is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S332.

  In step S332, the same processing as in step S122A is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S334.

  In step S334, based on the image determination data corresponding to the new defect addition image determined in step S313, it is determined whether or not the new defect addition image is a non-defective image. If YES at step S334, the process proceeds to step S340. On the other hand, if NO at step S334, the process proceeds to step S336.

  In step S336, the same processing as in step S130 is performed, and thus detailed description will not be repeated. Thereafter, the process proceeds to step S340.

  In step S340, a new additional teaching defect drawing process B is performed. In the new additional teaching defect rendering process B, if it is determined in step S334 that the new defect adding image is a non-defective image, image data to be displayed on the display unit 530 is processed by the control unit 510 and the VDP 532. The new defect addition image 800 is drawn in the VRAM 536 for storing (see FIG. 19A). Then, by the processing of the control unit 510 and the VDP 532, the same additional position for the new defect 720 as the true defect 720 of the original image 700 for teaching is placed on the center of gravity of the true defect 720 of the original image 700 for teaching calculated in step S320 on the VRAM 536. A defect 820 is drawn (see FIG. 19B).

  In addition, in the new additional teaching defect rendering process B, when it is determined in step S334 that the new defect addition image is a defect image, the control unit 510 and the VDP 532 perform the same process as in step S140. The teaching original image 700 is drawn in the VRAM 536. Further, by the processing of the controller 510 and the VDP 532, the additional teaching defect generated in step S133B is generated on the VRAM 536 at the drawing position of the new additional teaching defect determined in step S336. Draw the defect. Thereafter, the process proceeds to step S350.

  In step S350, control unit 510 determines whether or not the number of newly added source teaching defects determined in step S322 is equal to the number of newly added teaching defects drawn in step S340. If YES in step S350, the process proceeds to step S360. On the other hand, if NO at step S350, the process proceeds to step S352.

  In step S352, control unit 510 increments counter k by “1”. Thereafter, the process of step S332 is performed again.

  In step S360, control unit 510 causes storage unit 520 to store the image data (new defect addition teaching image data) stored in VRAM 536. Then, the new defect addition teaching image generation process B ends.

  A new defect addition teaching image is generated by the new defect addition teaching image generation process B described above.

  Referring to FIG. 19 again, in the present embodiment, the new defect addition image read in the process of step S312 is a new defect addition image 800 that is a non-defective image. Therefore, a new defect addition teaching image generation process is performed. By B, a new defect addition teaching image 800A is generated.

  As described above, in the present embodiment, when the first input teaching original image is a defect image, the teaching original image data is newly read out and used as a new defect adding image. If the new defect addition image is a non-defective image, the true defect of the first teaching original image is located at the position corresponding to the center of gravity of the true defect of the first teaching original image in the new defect adding image. A new defect addition teaching image to which the same new additional teaching defect is added is generated.

  Therefore, the new defect additional teaching image has an effect that it is possible to reproduce the defect area and contrast variation in the environment where the background is the same as the additional source teaching defect of the first teaching original image.

  In addition, when the image for new defect addition is a defect image having a true defect, and the center of gravity position of the defect for original addition of the image for new defect addition is the same as the drawing position of the defect for new additional teaching. In this case, since the additional source teaching defect and the new additional teaching defect overlap, the new additional teaching defect cannot be drawn on the new defect adding image.

  However, when the new defect addition image is a non-defective image, there is no problem as described above, and the effect of being able to draw the new additional teaching defect on the new defect adding image as the teaching original image. Play.

<Fourth embodiment>
Next, an image processing algorithm generation process for generating an image processing algorithm by causing a learning algorithm to learn a new defect addition teaching image as a teaching image generated by the processing of the first to third embodiments described above. explain.

  FIG. 20 is a block diagram showing an internal configuration of image processing algorithm generation apparatus 500C in the present embodiment.

  Referring to FIG. 20, image processing algorithm generation apparatus 500 </ b> C has data access to recording medium 555 </ b> C instead of recording medium 555 as compared with teaching image generation apparatus 500 of FIG. The difference is that the image processing algorithm generation program 180C is recorded instead of the teaching image generation program 180 and the image processing algorithm generation program 180C is recorded on the recording medium 555C. Since other configurations and functions are the same as those of teaching image generation apparatus 500, detailed description will not be repeated. An image processing algorithm is generated by the image processing algorithm generation apparatus 500C having the above configuration.

  Next, processing for generating an image processing algorithm by the image processing algorithm generation apparatus 500C will be described. In the present embodiment, the combination of image processing filters is optimized by using a genetic algorithm, which is one of the evolutionary methods, as a combination optimization problem for searching for combinations of image processing filters. That is, the optimum combination of image processing filters (local solution) is searched. An image processing algorithm which is a combination of these image processing filters is expressed by one individual genetic algorithm. Hereinafter, the image processing algorithm is also referred to as an individual.

FIG. 21 is a flowchart of image processing algorithm generation processing.
Referring to FIG. 21, in step S400, control unit 510 sets initial parameters. Although details will be described later, the initial parameters include the number m of the N generations, the crossover probability Pc, the mutation probability Pm, the termination condition, the number of excellent individuals g1 added to the N + 1 generation, and the number of randomly generated individuals added to the N + 1 generation g2. , Probabilistic generation number g3 considering N generation best individual added to N + 1 generation, Probability change rate at probability generation considering N generation best individual added to N + 1 generation, change of probability change rate Rate, the maximum value L of the image processing filter length of one individual, and the like. In the present embodiment, it is assumed that the number m of N generations is set to “100”, for example.

  Note that g1 + g2 + g3 ≦ m holds for the above parameters. The number of excellent individuals g1, the number of randomly generated individuals g2, and the number of randomly generated individuals g3 are desirably 5% or less with respect to the number of individuals m.

  The end condition in the present embodiment is a condition that the evaluation value of the best individual is a predetermined value (for example, 1000) or more, or a condition that the number of times of repeating the generation change process described later is a predetermined number (for example, 100). It is. Thereafter, the process of step S401 is performed.

  In step S401, control unit 510 randomly generates m first (one generation) image processing algorithms. In the present embodiment, the image processing algorithm is based on a combination of a plurality of types of image processing filters prepared in advance.

  FIG. 22 shows the filter table T300. The filter table T300 indicates a correspondence between a filter ID for specifying the type of the image processing filter and the image processing filter. The filter IDs “1”, “2”, “3”, “4”, “5”, “6” respectively include a smoothing filter, an intermediate value filter, a maximum value filter, a minimum value filter, and a Sobel. A filter and a Laplacian filter are associated with each other.

  For example, the smoothing filter is a filter that performs a process of replacing the density value of the target pixel with the average value of the density values of the target pixel and pixels in the vicinity thereof. In addition, as a parameter of the smoothing filter, for example, there is a parameter that defines a neighboring region. For example, when the value of the parameter is “3”, the filter size is defined as 3 pixels × 3 pixels. In this case, the target pixel and 8 pixels in the vicinity thereof are targeted for filtering. The filter size used by the smoothing filter, the intermediate value filter, and the maximum value filter is not limited to 3 pixels × 3 pixels, and may be any size (for example, 5 pixels × 5 pixels).

  The filter used in the present embodiment is not limited to the above filter, and may be another filter (for example, a Gaussian filter). In the present embodiment, the density value of the pixel is a value in the range of “0” to “255”. If the density value is “0”, the corresponding pixel is assumed to be black. If the density value is “255”, the corresponding pixel is white. The image processing algorithm is expressed by a combination of various image processing modules (filters) and parameter values possessed by each image processing module.

  Referring to FIG. 21 again, in step S401, control unit 510 generates an individual obtained by combining the filters based on the maximum value L of the image processing filter length of one individual. Here, the maximum value L indicates the maximum number of filters to be combined. In the present embodiment, it is assumed that the maximum value L is, for example, “8”. The maximum value L is not limited to “8” and may be an arbitrary value (for example, “10”). The maximum value L is a value empirically determined by the user, and it is desirable that the maximum value L is set larger as the difficulty of processing by the individual is higher.

  The control unit 510 randomly determines the number of filters in the range of “1” to “8” for each individual to be generated. Then, based on the number of filters determined for each individual to be generated, the control unit 510 randomly selects an image processing filter to be used from among a plurality of types of filters (six types in FIG. 22) without preventing duplication. To do.

FIG. 23 is a diagram showing an individual generated in the present embodiment.
FIG. 23A shows m individuals I (N) _1 to I (N) _m randomly generated in step S401. By step S401, the number of filters used by each of the m individuals I (N) _1 to I (N) _m becomes random. The number constituting each individual indicates a filter ID. For example, in the individual I (N) _1, the processing target image is sequentially processed by the filters corresponding to the filter IDs “1”, “3”, “1”, “6”, “5”. That is, the filter ID corresponding to the leftmost side of the individual I (N) _1 is the first filter used for the processing target image. The filter ID corresponding to the rightmost side of the individual I (N) _1 is the last filter used for the processing target image. Hereinafter, an image obtained by performing image processing of the individual I (N) _k (natural number) on the processing target image is also referred to as a post-processing image.

  Referring to FIG. 21 again, when the process of step S401 ends, the process proceeds to step S402.

  In step S 402, control unit 510 selects processing target image data as a teaching image from teaching image database 182 stored in storage unit 520, and reads processing target image data from storage unit 520. It is assumed that the processing target image to be read is, for example, a new defect addition teaching image generated by the processing of the second embodiment among the first to third embodiments described above. The read processing target image may be a new defect addition teaching image generated by the processing of the first and third embodiments. The teaching image database 182 is a database including a plurality of types of teaching image data, each of which is different.

  In the present embodiment, the teaching image is classified into a processing target image and a teaching target image. The target image for teaching is an image for which it is desired that an image obtained by performing image processing with an image processing algorithm on the processing target image matches or approximates. It can be said that the image recognition rate of the image processing algorithm is higher as the image obtained by performing the image processing on the processing target image is closer to the teaching target image.

  Further, the teaching target image is not limited to a binary image, and may be, for example, a ternary image having density values of “0”, “128”, and “255”. In this case, for example, only pixels having a density value near “128” are extracted. The teaching target image is, for example, an image in which a desired extraction region is arranged.

  FIG. 24 is a diagram showing a processing target image 700B and a teaching target image 700C in the present embodiment.

  FIG. 24A shows a processing target image 700B. The processing target image 700B includes true defects 720, 722, and 724 and pseudo defects 710, 712, and 714.

  FIG. 24B is a diagram showing a teaching target image 700C. The teaching target image 700C has desired extraction regions 720C, 722C, and 724C. In the teaching target image 700C, only the pixels in the extraction areas 720C, 722C, and 724C are white, and the pixels other than the extraction areas 720C, 722C, and 724C are black. Note that the processing target image 700B and the teaching target image 700C corresponding to the processing target image are associated with each other and stored in the storage unit 520.

  Referring to FIG. 21 again, when the process of step S402 ends, the process proceeds to step S403A.

  In step S403A, control unit 510 sets generation number N representing a generation to “1” in individual I (N) _m. The generation number N is incremented by “1” every time a generation change process described later is performed. Thereafter, the process proceeds to step S403B.

  In step S403B, control unit 510 sets counter k used to execute a process to be described later to “1” for each of m individuals I (N) _1 to I (N) _m. Thereafter, the process proceeds to step S404.

  In step S404, the control unit 510 performs image processing on the kth individual I (N) _k among the m individuals I (N) _1 to I (N) _m generated in step S401. To generate an image after processing.

  For example, when k = 1, image processing by the individual I (N) _1 is performed on the processing target image. In this case, the processing target image is sequentially processed by the filters corresponding to the filter IDs “1”, “3”, “1”, “6”, “5”. An image obtained by performing image processing on the processing target image with the individual I (N) _1 is a processed image of the individual I (N) _1. When the process of step S404 ends, the process proceeds to step S405.

  In step S405, control unit 510 calculates an evaluation value of the individual (image processing algorithm) from the processed image. In the present embodiment, the absolute values of the differences in density values between the pixels of the teaching target image and the processed image based on the data included in the teaching image data read out in step S402 are all set. Find pixels. Then, a value obtained by adding the absolute values obtained for each pixel for all the pixels (hereinafter also referred to as an image difference value) is obtained and subtracted from a predetermined value (for example, 2000). In the present embodiment, the value obtained by the subtraction is used as the evaluation value of the image processing algorithm.

  Therefore, the smaller the image difference value, that is, the larger the evaluation value, the closer the processed image generated by the corresponding individual (image processing algorithm) is to the teaching target image. That is, it can be said that the individual (image processing algorithm) has a high recognition rate for recognizing the processing target image as the teaching target image. That is, it can be said that the larger the evaluation value of an individual (image processing algorithm), the better (higher recognition rate) image processing algorithm. When the process of step S405 ends, the process proceeds to step S406.

  In step S406, control unit 510 increments counter k by 1. Thereafter, the process proceeds to step S407.

  In step S407, control unit 510 determines whether counter k is greater than the number m of individuals. If YES in step S407, the process proceeds to step S408. On the other hand, if NO at step S407, the process at step S404 is performed again. That is, when the calculation of evaluation values for all m individuals is completed, the process proceeds to step S408.

  In step S408, control unit 510 determines the best individual (image processing algorithm) from all m individuals. In this case, the individual with the largest calculated evaluation value (image processing algorithm) is the best individual. Thereafter, the process proceeds to step S409.

  In step S409, control unit 510 determines whether or not the above-described termination condition is satisfied. If YES in step S409, the image processing algorithm generation process ends. The image processing algorithm generation process ends when, for example, the evaluation value of the best individual (image processing algorithm) of a generation is 1000 or more. Or, the generation number N is a value of 100 or more. That is, this is a case where the generation change process described later is performed 100 times. When the generation change process is performed 100 times, the image processing algorithm generation process ends even if the evaluation value of the best individual (image processing algorithm) of that generation is less than 1000.

  All m individuals (image processing algorithm) in the generation when the image processing algorithm generation processing is completed (hereinafter also referred to as the final generation) or the best individual of the final generation determined in step S408 (image processing) Algorithm) is the final individual (image processing algorithm), that is, the final image processing algorithm.

  On the other hand, if NO at step S409, the process at step S410 is performed.

  In step S410, next-generation image processing algorithm generation processing is performed using an evolutionary method. Note that the next-generation image processing algorithm generation processing described below is called generation change processing or genetic operation in an evolutionary method. Also in the genetic algorithm, the next generation image processing algorithm generation processing described below is called generation change processing or genetic operation.

FIG. 25 is a flowchart of next-generation image processing algorithm generation processing.
Referring to FIG. 25, in step S510, control unit 510 ranks generation number N, that is, m individuals (image processing algorithm) of N generations based on corresponding evaluation values. In the present embodiment, the individual with the highest evaluation value is the best (first) individual, and the individual with the lowest evaluation value is the mth individual. If there are a plurality of individuals having the same evaluation value, the order between the plurality of individuals is arbitrary. Thereafter, the process proceeds to step S520.

  In step S520, among the m individuals ranked in step S510, the first to g1 individuals are added to the N + 1 generation without change based on the number of excellent individuals g1 set in step S400 described above. To do.

  Here, the number of excellent individuals g1 is the number of individuals to be added to the (N + 1) generation without being changed among the m individuals of the N generations. In the present embodiment, it is assumed that the number of excellent individuals g1 is “1”. That is, in the present embodiment, only the first individual is added to the N + 1 generation. Specifically, control unit 510 causes storage unit 520 to store the first individual (image processing algorithm). It is assumed that there are no individuals in the N + 1 generation at first. The number of excellent individuals g1 is not limited to “1” and may be an arbitrary value (for example, “2”).

  Referring to FIG. 23 again, FIG. 23 (B) is a diagram showing an N + 1 generation individual. In the present embodiment, it is assumed that the number m of individuals does not change when an N + 1 generation individual is generated. Through the process of step S520, the N generations of individuals I (N) _1 are not changed, that is, the filters are not changed, and are added to the N + 1 generation to become individuals I (N + 1) _1.

  Referring to FIG. 25 again, when the process of step S520 ends, the process proceeds to step S530.

  In step S530, a stochastic generation process is performed in consideration of the best image processing algorithm of N generations. In the probabilistic generation process, the control unit 510 ranks the m individuals performed in step S510, sets the probability generation number g3, the probability variation rate, and the probability variation rate change rate set in step S400 described above. Based on this, g3 individuals are generated and added to the N + 1 generation.

  In the individual generation method in the stochastic generation process, the selection probability of selecting a filter from a plurality of types of filters in the N + 1 generation individual is changed based on the filter used in the best individual of the N generations.

  Here, the number of probabilistic generations g3 is the number of individuals considering the best individual of N generations to be added to the N + 1 generation. The probability variation ratio Pf is a probability at the time of probabilistic generation in consideration of the best individual of N generations added to the N + 1 generation. The probabilistic generation number g3, the probability variation rate, and the change rate are values determined by the user empirically.

  In the present embodiment, it is assumed that the number of probabilistically generated individuals g3 is “2”. The probabilistic generation number g3 is not limited to “2”, and may be an arbitrary value.

  Referring to FIG. 23B again, in step S530, the individual I (N + 1) _2 and the individual I (N + 1) _3 are added to the N + 1 generation.

Referring to FIG. 25 again, after the process of step S530, the process proceeds to step S540.
In step S540, g2 individuals are randomly generated based on the number g2 of randomly generated individuals set in step S400 described above and added to the N + 1 generation by the same method as in step S401. Here, the randomly generated number of individuals g2 is the number of individuals that are randomly generated and added to the N + 1 generation. In the present embodiment, it is assumed that the number of randomly generated individuals g2 is “1”. Note that the number of randomly generated individuals g2 is not limited to “1” and may be an arbitrary value (for example, “2”).

  Referring to FIG. 23B again, in step S540, the individual I (N + 1) _4 is added to the N + 1 generation. Note that the individual I (N + 1) _4 may be the same as the individual I (N) _4.

  Referring to FIG. 25 again, when the process of step S540 is completed, the process proceeds to step S550.

  In step S550, m- (g1 + g2 + g3) individuals are generated from m (100) individuals of N generations by crossover, which is one of the genetic operations. That is, 100− (1 + 1 + 2) = 96 individuals are generated by crossover from m (100) individuals of N generations.

  FIG. 26 is a diagram illustrating an example of an individual selection method for crossover or mutation in the present embodiment.

  Referring to FIG. 26, first, control unit 510 randomly selects 10 pieces, which is 10% of 100 pieces, from 100 individuals (FIG. 26B). Then, control unit 510 selects the individual with the highest evaluation value as the best individual among the 10 individuals selected at random (FIG. 26C). Control unit 510 repeats the above process twice to randomly select two individuals (hereinafter, also referred to as two crossover targets) from 100 individuals.

  Then, the control unit 510 crosses two crossing target individuals based on the crossing probability Pc described above, and generates two individuals. Here, the crossover probability Pc is the probability of crossing two crossover target individuals. In the present embodiment, it is assumed that the crossover probability Pc is 0.9 (90%). Note that the crossover probability Pc is not limited to 0.9. Crossover is performed by the method described below.

FIG. 27 is a diagram for explaining crossover and mutation.
FIG. 27A and FIG. 27B are diagrams for explaining crossover.

  Referring to FIG. 27A and FIG. 27B, crossing means cutting two adjacent components at an arbitrary position (hereinafter, the crossing point) and exchanging the cut ones with each other. By doing so, two new individuals of the (N + 1) th generation are generated.

  By repeating the above crossover process, 96 individuals are generated from m (100) individuals of N generations.

  Referring to FIG. 25 again, when the process of step S550 ends, the process proceeds to step S560.

  In step S560, the 96 individuals generated in step S550 (hereinafter also referred to as mutation target individuals) are mutated as one of the genetic operations based on the mutation probability Pm. (In the following, also referred to as mutation treatment). Here, the mutation probability Pm is a probability of performing mutation on 96 mutation target individuals. In the present embodiment, it is assumed that the mutation probability Pm is 0.05 (5%). The mutation probability Pm is not limited to 0.05. In the present embodiment, for mutation, control unit 510 performs the method described below with a probability of 5% for each of the 96 mutation target individuals.

Referring to FIG. 27 again, FIG. 27C is a diagram for explaining the mutation.
Referring to FIG. 27C, mutation means that an arbitrary position (mutation part) of one individual of the Nth generation is selected, and the type of individual (image processing filter) corresponding to the mutation part is selected. A new individual of the (N + 1) th generation is generated by changing to an arbitrary individual (image processing filter) at random. There may be a plurality of mutation parts.

  Then, control unit 510 adds 96 individuals that have been subjected to mutation processing to the N + 1 generation. Specifically, control unit 510 causes storage unit 520 to store the generated 96 individuals (image processing algorithm). Through the above processing, the number of individuals of the (N + 1) th generation is the same as the number of individuals of the Nth generation. That is, the number of individuals (image processing algorithms) stored in the storage unit 520 is 100. Then, this next-generation image processing algorithm generation process ends, the process returns to the image processing algorithm generation process of FIG. 21, and the process proceeds to step S412.

  Note that the order of the processes in steps S520, S530, S540, S550, and S560 in the next-generation image processing algorithm generation process in the present embodiment is not limited to the order described above, and is arbitrary. For example, the process of step S530 may be performed prior to the process of step S520.

  In step S412, control unit 510 increments generation number N by one. Thereafter, the process of step S403B is performed again.

  Then, the processes of steps S403B, S404, S405, S406, S407, S408, S410, and S412 are repeated until it is determined in step S409 that the end condition is satisfied. Through the above processing, m individuals (image processing algorithms) are changed to excellent (high recognition rate) m individuals (image processing algorithms) as generations change. When the end condition is satisfied in step S409, the storage unit 520 stores 100 excellent individuals (image processing algorithms).

  As described above, in this embodiment, a genetic algorithm is used as an optimization method. However, other evolutionary methods such as genetic programming and evolutionary algorithm may be used, and other than evolutionary methods such as a neural network. An optimization method may be used.

  Further, the image processing filter illustrated in FIG. 22 is an example, and noise removal, binarization processing, and the like may be included in the image processing filter.

  FIG. 28 is a functional block diagram of an image processing algorithm generation device 500C in the present embodiment.

  Referring to FIG. 28, image processing algorithm generation apparatus 500C operates as an image processing algorithm management unit. An image data input unit 600 and a display unit 530 are connected to the image processing algorithm generation apparatus 500C.

The image processing algorithm generation device 500C includes a control unit 510.
The control unit 510 operates as an initial parameter setting unit 511C, a target image setting unit 512C, an image processing algorithm generation unit 513C, an image data conversion unit 514C, an image processing algorithm performance evaluation unit 515C, and an image processing algorithm generation completion determination unit 516C. The teaching image database 182 is stored in the storage unit 520. The teaching image database 182 includes the processing target image input from the image data input unit 600.

  The initial parameter setting unit 511C has a function of changing an initial parameter based on an instruction from the user using an input device such as a mouse 524 or a keyboard 544 when the user wants to change an initial parameter set in advance. . The initial parameter setting unit 511C has a function of performing the process of step S400 described above.

  The target image setting unit 512C refers to the processing target image of the teaching image data constituting the teaching image database 182 on the display unit 530, and the target image setting unit 512C receives information from the user using the input device such as the mouse 524 or the keyboard 544. Based on the instruction, it has a function of generating a target image for teaching in which a desired extraction area is arranged.

  The image processing algorithm generation unit 513 has a function of performing the processes of steps S401 and S410 described above.

The image data conversion unit 514C has a function of performing the process of step S404 described above.
The image processing algorithm performance evaluation unit 515C has a function of performing the processes of steps S405 and S408 described above.

  The image processing algorithm generation completion determination unit 516C has a function of performing the process of step S409 described above.

  When the image processing algorithm generation completion determination unit 516C determines that generation of a good individual (image processing algorithm) is complete, the storage unit 520 is an image processing algorithm storage unit that stores the good individual (image processing algorithm). Operate.

  FIG. 29 shows simulation results obtained by measuring the time required to generate the 100th generation image processing algorithm in each of the conventional method and the present embodiment. The simulation conditions were such that the image size to be used was 640 pixels wide and 480 pixels vertical, the number m of individuals of each generation was 100, and the maximum value L of the filter to be used was 8.

  Referring to FIG. 29, in the conventional method, nine images in which the defect feature amount is changed are created for one processing target image (teaching image), and a total of 10 processing target images are recorded. Used in the image processing algorithm generation processing of the embodiment.

  In the present embodiment, a new defect additional teaching image created by adding nine new additional teaching defects generated based on one additional source teaching defect by the method of the second embodiment. Was used in the image processing algorithm generation processing of the present embodiment.

  In the conventional method, the time required to repeat the generation change process 100 times is 4269 seconds, but in this embodiment, the time can be significantly reduced to 801 seconds.

  As described above, in this embodiment, one new defect addition teaching image created by adding a plurality of new additional teaching defects generated based on one additional source teaching defect is obtained. By generating an image processing algorithm as a processing target image, it is possible to reduce the number of times image processing is performed, compared to the conventional method of generating an image processing algorithm using a plurality of types of processing target images.

  Therefore, the time required for generating the image processing algorithm can be shortened.

  Also, in this embodiment, by using an evolutionary method that is a global search method, it is possible to efficiently search even when the number of image processing filter combinations is enormous, and it is necessary to generate an image processing algorithm. There is an effect that the time can be shortened.

<Fifth embodiment>
Next, processing for performing an image inspection using the image processing algorithm generated in the fourth embodiment will be described.

  FIG. 30 is a block diagram showing an internal configuration of image inspection apparatus 500D in the present embodiment.

  Referring to FIG. 30, image inspection apparatus 500 </ b> D performs data access to recording medium 555 </ b> D instead of recording medium 555, as compared with teaching image generation apparatus 500 of FIG. An image inspection program 180D is recorded instead of the image generation program 180, an image inspection program 180D is recorded on the recording medium 555D, and an image processing algorithm database 184 is further recorded in the storage unit 520. The point is different. Since other configurations and functions are the same as those of teaching image generation apparatus 500, detailed description will not be repeated.

  The image processing algorithm database 184 is a database composed of a plurality of image processing algorithms generated in the fourth embodiment. Here, the image processing algorithm is composed of one or more image processing filters. Image inspection processing is performed by the image inspection apparatus 500D having the above configuration.

Next, the image inspection process will be described.
FIG. 31 is a flowchart of the image inspection process.

  Referring to FIG. 31, in step S601, an initial parameter setting process is performed. In the initial parameter setting process, control unit 510 sets initial parameters. The initial parameters include the ID of the image processing algorithm selected in step S603, which will be described later, and a defect area threshold value for determining whether the defect is a true defect or a pseudo defect. The inspection object 650 having a defect with an area equal to or larger than the threshold is a defective product, and the inspection object 650 having no defect with an area larger than the threshold is a non-defective product.

  In particular, by setting this threshold value to 1, if there is at least one pixel having a density value of 255 (white) in a binarized image that is an output of an image processing algorithm application result described later, the inspection object 650 is defective. Can be determined. Thereafter, the process proceeds to step S602.

  In step S <b> 602, an inspection target image is input by imaging the inspection target 650 using the image data input unit 600. When performing offline image inspection, an image is input by selecting an inspection target image as a processing target image from the teaching image database 182. Thereafter, the process proceeds to step S603.

  In step S603, control unit 510 selects an image processing algorithm having the ID specified in step S601 from image processing algorithm database 184 stored in storage unit 520, and stores the selected image processing algorithm from storage unit 520. read out.

  Note that an output image obtained as a result of applying the image processing algorithm to a certain inspection target image is a binary image having a density value of 0 or 255. In particular, in the present embodiment, it is assumed that only the density value of the defect candidate pixel is 255 (white). Thereafter, the process proceeds to step S604.

  In step S604, control unit 510 performs the image processing algorithm read in step S603 on the inspection target image input in step S602, and outputs a binary image as an image processing result. Thereafter, the process proceeds to step S605.

  In step S605, control unit 510 calculates feature amounts of all defect candidates using a labeling method for the binarized image obtained in step S604. Here, the feature amount is, for example, an area. Thereafter, the process proceeds to step S606.

  In step S606, control unit 510 uses the defect area threshold for determining whether inspection object 650 set in step S601 is a non-defective product or a defective product to determine whether each defect candidate is a true defect or a pseudo defect. It is determined whether it is. Based on the determination, control unit 510 determines that inspection object 650 having at least one true defect is a defective product. Hereinafter, the determination result is also referred to as a non-defective product / defective product determination result. Thereafter, the process proceeds to step S607.

  In step S607, control unit 510 and VDP 532 output the non-defective / defective product determination result in step S606 to display unit 530.

Thus, the image inspection process ends.
FIG. 32 is a functional block diagram of an image inspection apparatus 500D in the present embodiment.

  Referring to FIG. 32, image inspection apparatus 500D operates as an image inspection management unit. An image data input unit 600 and a display unit 530 are connected to the image inspection apparatus 500D.

  Image inspection apparatus 500D includes a control unit 510, an image inspection result output unit 510D, and a storage unit 520.

  The control unit 510 operates as an initial parameter setting unit 511D, an image processing algorithm selection unit 512D, an image data conversion unit 513D, a defect candidate feature amount calculation unit 514D, and a non-defective / defective product determination unit 515D.

  The storage unit 520 stores a teaching image database 182 and an image processing algorithm database 184. The teaching image database 182 includes the processing target image input from the image data input unit 600.

Image inspection result output unit 510D operates control unit 510 and VDP 532.
The image inspection apparatus 500D acquires the inspection target image from the image data input unit 600. When performing offline image inspection, at least one or more inspection object image data acquired by the image data input unit 600 before the inspection is stored in the storage unit 520, and the teaching image database 182 of the storage unit 520 is stored at the time of inspection. The image data to be inspected is acquired.

  The initial parameter setting unit 511D provides a function of changing an initial parameter by using an input device such as a mouse 542 or a keyboard 544 when the user wants to change an initial parameter set in advance. The initial parameter setting unit 511D has a function of performing the process of step S601 in FIG.

  The image processing algorithm selection unit 512D has a function of performing the process of step S603. That is, a desired image processing algorithm is selected from a plurality of image processing algorithms stored in the image processing algorithm database 184.

The image data conversion unit 513D has a function of performing the process of step S604.
The defect candidate feature amount calculation unit 514D has a function of performing the process of step S605.

The non-defective product / defective product determination unit 515D has a function of performing the process of step S606.
The image inspection result output unit 510D has a function of performing the process of step S607.

  Next, a case where image inspection processing using two different image processing algorithms is performed will be described.

  FIG. 33 is a diagram showing two different processing target images used to generate the image processing algorithm selected in step S603.

  FIG. 33A shows a processing target image 900. The processing target image 900 is an image obtained by imaging the inspection target object 650. The processing target image 900 includes a true defect 820.

  FIG. 33B shows a processing target image 900A. The processing target image 900A is an image generated using the processing target image 900 by the methods of the first and second embodiments. The processing target image 900A is an image in which a new additional teaching defect 824 in which the area and contrast of the true defect 820 are reduced by 25% is generated as a true defect and added.

  FIG. 34 is a table T400 showing the results of image inspection processing using two different image processing algorithms.

  Referring to FIG. 34, image processing algorithm A is the best image processing algorithm generated in the image processing algorithm generation processing of the fourth embodiment, with the processing target image as processing target image 900.

  The image processing algorithm B is the best image processing algorithm generated in the image processing algorithm generation processing of the fourth embodiment with the processing target image as the processing target image 900A.

  ○ indicates that when the inspection object image obtained by imaging the inspection object 650 is a defect image including a true defect, the inspection object 650 is correctly determined to be defective in step S606. Show. That is, when the defect image includes a true defect, it indicates that the defect candidate included in the defect image is correctly determined as a true defect.

  X: When the inspection object image obtained by imaging the inspection object 650 is a defect image including a true defect, the inspection object 650 could not be correctly determined as a defective product in step S606. Indicates. That is, if the defect image includes a true defect, it indicates that the defect candidate included in the defect image has not been correctly determined as a true defect.

  FIG. 35 is a diagram showing an inspection target image 950 as an example used in the image inspection processing using two different image processing algorithms.

  Referring to FIG. 35, inspection object image 950 includes a true defect 952. The true defect 952 is a defect in which the area or contrast of the true defect 820 of the processing target image 900 in FIG. The density distribution of the background image of the inspection target image 950 and the background image of the processing target image 900 are the same.

5 to 35 in the row of the variation feature amount respectively indicate% in which the corresponding feature amount is reduced.
First, by using the processing target image 900, a plurality of inspection target images in which only the area of the true defect 820 of the processing target image 900 is reduced by 5 to 35% are generated, and a step is performed for each of the plurality of inspection target images. In step S604, the image processing algorithm A is applied to perform image inspection processing. In the image inspection process, the inspection target image was correctly determined as a defect image even for a true defect in which the area of the true defect 820 was reduced to 15%. That is, it was correctly determined that the inspection object 650 is a defective product.

  Further, using the processing target image 900, a plurality of inspection target images in which only the contrast of the true defect 820 of the processing target image 900 is reduced by 5 to 35% are generated, and a step is performed for each of the plurality of inspection target images. In step S604, the image processing algorithm A is applied to perform image inspection processing. In the image inspection processing, the inspection target image was correctly determined as a defect image even for a true defect in which the contrast of the true defect 820 was reduced to 20%. That is, it was correctly determined that the inspection object 650 is a defective product.

  Further, by using the processing target image 900, a plurality of inspection target images in which only the area of the true defect 820 of the processing target image 900 is reduced by 5 to 35% are generated, and a step is performed for each of the plurality of inspection target images. In step S604, the image processing algorithm B is applied to perform image inspection processing. In the image inspection process, the inspection target image was correctly determined as a defect image even for a true defect in which the area of the true defect 820 was reduced to 25%. That is, it was correctly determined that the inspection object 650 is a defective product.

  Also, using the processing target image 900, a plurality of inspection target images in which only the contrast of the true defect 820 of the processing target image 900 is reduced by 5 to 35% are generated, and a step is performed for each of the plurality of inspection target images. In step S604, the image processing algorithm B is applied to perform image inspection processing. In the image inspection processing, the inspection target image was correctly determined as a defect image even for a true defect in which the contrast of the true defect 820 was reduced to 35%. That is, it was correctly determined that the inspection object 650 is a defective product.

  As described above, the image inspection processing using the image processing algorithm B generated using the image with the defect added (processing target image 900A) uses the image without the defect added (processing target image 900). From the image inspection process using the image processing algorithm A generated in this way, the defect image among the plurality of inspection object images in which the feature amount of the true defect 820 of the processing object image 900 is changed using the processing object image 900. There were many images that could be judged correctly.

  As described above, in the present embodiment, the image processing algorithm B generated in the fourth embodiment using the processing target image generated in the first and second embodiments is used in the image inspection process. I use it.

  Therefore, the image processing algorithm B used in the image inspection process also learns the feature of a new defect whose feature amount of the defect existing on the original image for teaching has changed, so that the illumination conditions and camera parameters are different. Even if the environment changes, there is an effect that a highly reliable inspection is possible.

  In the present embodiment, the defect candidate area is used as a non-defective / defective product determination criterion, but the present embodiment is not limited to this. The criteria for determining good / defective products may be anything as long as they are criteria (for example, the shape of a defect candidate) that can determine whether the inspection object is a good product or a defective product.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

  The present invention is used for an image inspection apparatus, an image conversion apparatus, an image recognition apparatus, a pattern matching apparatus, a positioning apparatus using an image, an autonomous robot with an image recognition function, an industrial robot with an image recognition function, and the like.

It is a block diagram which shows the internal structure of the image generation apparatus for teaching in this Embodiment. It is a block diagram which shows the internal structure of an image data input part. It is a figure which shows the image for a teaching as an example. It is a figure which shows the original image for teaching and the target image for teaching as an example. It is a flowchart of the image generation process for new defect addition teaching. The teaching original image, and the additional original teaching defect designation image generated based on the teaching original image are shown. It is a defect area setting image in which a defect area is set for a teaching original image. It is a figure which shows the defect list table for newly added teachings as an example. It is a flowchart of new addition defect drawing position determination processing. A new defect additional teaching image in which a new additional teaching defect is arranged in the teaching original image is shown. It is a defect area setting image in which a defect area is set for a teaching original image. It is a functional block diagram of the teaching image generation apparatus in the present embodiment. It is a block diagram which shows the internal structure of the image generation apparatus for teaching in this Embodiment. It is a flowchart of the image generation process A for new defect addition teaching. It is a figure which shows the defect list table for newly added teachings as an example. It is a figure which shows the image for a new defect addition teaching produced | generated in this Embodiment. It is a block diagram which shows the internal structure of the image generation apparatus for teaching in this Embodiment. It is a flowchart of the image generation process B for new defect addition teaching. The image for new defect addition and the image for new defect addition teaching in this Embodiment are shown. It is a block diagram which shows the internal structure of the image processing algorithm production | generation apparatus in this Embodiment. It is a flowchart of an image processing algorithm generation process. A filter table is shown. It is a figure which shows the individual | organism | solid produced | generated in this Embodiment. It is a figure which shows the process target image and the target image for teaching in this Embodiment. It is a flowchart of a next-generation image processing algorithm generation process. It is a figure which shows an example of the selection method of the individual for the crossover or mutation in this Embodiment. It is a figure for demonstrating crossing and a mutation. It is a functional block diagram of the image processing algorithm production | generation apparatus in this Embodiment. It is the simulation result which measured the time required in order to produce | generate the image processing algorithm of the 100th generation in each of the conventional method and this Embodiment. It is a block diagram which shows the internal structure of the image inspection apparatus in this Embodiment. It is a flowchart of an image inspection process. It is a functional block diagram of the image inspection apparatus in the present embodiment. It is a figure which shows two different process target images used in order to produce | generate an image processing algorithm. It is Table T400 which shows the result of having performed the image test | inspection process which each uses two different image processing algorithms. It is a figure which shows the test object image as an example used by the image test | inspection process which each uses two different image processing algorithms.

Explanation of symbols

  180, 180A, 180B Teaching image generation program, 180C image processing algorithm generation program, 180D image inspection program, 182 Teaching image database, 184 image processing algorithm database, 500, 500A, 500B Teaching image generation device, 500C image processing algorithm Generation device, 500D image inspection device, 510 control unit, 520 storage unit, 530 display unit, 532 VDP, 555, 555A, 555B, 555C recording medium, 600 image data input unit.

Claims (38)

A teaching image generation method for generating a teaching image having an inspection region for processing by an algorithm that learns a target process in a self-developed manner in a teaching process,
Inputting a teaching image having the inspection area;
Designating the inspection area;
Calculating a feature amount of an inspection region image in the designated inspection region;
Calculating a position where the image is arranged in the teaching image;
Generating a change inspection region image in which the feature amount of the inspection region image is changed;
Generating a new defect addition teaching image in which the change inspection area image is arranged at a position where the calculated image is arranged. Further comprising receiving a user instruction,
The teaching image generation method according to claim 1, wherein the step of designating the inspection area designates the inspection area based on the received user instruction.   The teaching image generation method according to claim 1, wherein a position where the image is arranged is a position that does not overlap the inspection region image. The step of calculating the feature amount of the inspection area image includes
Calculating a center of gravity position of the inspection region image,
The step of calculating the position where the image is to be arranged is calculated by searching for the position where the image is to be arranged from the vicinity around the calculated gravity center position of the inspection region image. The teaching image generating method according to claim 1. Further comprising determining whether the inspection area is an area to be extracted;
The step of generating the change inspection region image generates the change inspection region image in which the area of the inspection region image is reduced when it is determined that the inspection region is the region to be extracted. The teaching image generation method according to claim 4. Further comprising determining whether the inspection area is an area to be extracted;
The step of generating the change inspection region image generates the change inspection region image in which the area of the inspection region image is enlarged when it is determined that the inspection region is not the region to be extracted. The teaching image generation method according to claim 4. Further comprising determining whether the inspection area is an area to be extracted;
The step of calculating the feature amount of the inspection area image includes
Calculating a center of gravity position of the inspection region image,
In the step of inputting the teaching image, when it is determined that the inspection region is the region to be extracted, a new teaching image having the inspection region is newly input.
The step of generating the new defect addition teaching image generates a new defect addition teaching image in which the change inspection area image is arranged at a position corresponding to the center of gravity position of the inspection area image in the new teaching image. The teaching image generation method according to any one of claims 1 to 3. Further comprising determining whether the inspection area is an area to be extracted;
The step of calculating the feature amount of the inspection area image includes
Calculating a center of gravity position of the inspection region image,
In the step of inputting the teaching image, when it is determined that the inspection region is the region to be extracted, a new teaching image having the inspection region is newly input.
Determining whether the inspection area of the new teaching image is the area to be extracted;
In the step of generating the new defect addition teaching image, when it is determined that the inspection area of the new teaching image is not the area to be extracted, the position of the center of gravity of the inspection area image in the new teaching image 4. The teaching image generation method according to claim 1, wherein a new defect addition teaching image in which the change inspection area image is arranged at a position corresponding to is generated. 5.   The step of generating the change inspection region image generates the change inspection region image in which the area of the inspection region image is reduced when it is determined that the inspection region is the region to be extracted. Alternatively, the teaching image generation method according to claim 8.   The step of generating the change inspection region image generates the change inspection region image obtained by enlarging the area of the inspection region image when it is determined that the inspection region is not the region to be extracted. The teaching image generation method according to claim 8. Setting the first and second inspection surrounding areas around the inspection area image and the change inspection area image, respectively;
The first density value of the inspection area image, the second density value of the first background image in the background area other than the inspection area image, which is an area in the first inspection surrounding area, and in the second inspection surrounding area And calculating a fourth density value of the change inspection area image based on a third density value of a second background image of a background area other than the change inspection area image. The teaching image generation method according to claim 1. The first density value is an average density value of all pixels of the inspection area image;
The second density value is an average density value of all pixels of the first background image,
The third density value is an average density value of all pixels of the second background image;
Making the inspection area image the same size and shape as the change inspection area image;
The step of calculating the fourth density value of the change inspection area image includes setting the density values of all pixels of the change inspection area image to all pixels of the inspection area image having the same size and shape as the change inspection area image. A process for setting each density value,
The difference between the average value of the density values of all the pixels of the set change inspection region image and the third density value, and the difference value between the first density value and the second density value A step of adding or subtracting the absolute value to the density value of all the pixels of the set change inspection region image so that the absolute value of the value is not less than the predetermined value when the absolute value is not less than the predetermined value. The teaching image generation method according to claim 11, further comprising: A teaching image generating device for generating a teaching image having an inspection region for processing by an algorithm that learns a target process in a self-developed manner in a teaching process,
An image input unit for inputting a teaching image having the inspection region;
An inspection area designating unit for designating the inspection area;
A feature amount calculating unit for calculating a feature amount of an inspection region image in the designated inspection region;
A position calculation unit for calculating a position where the image is arranged in the teaching image;
A change inspection region image generation unit that generates a change inspection region image in which the feature amount of the inspection region image is changed;
An teaching image generation apparatus comprising: an image generation unit that generates a new defect addition teaching image in which the change inspection area image is arranged at a position where the calculated image is arranged. It further includes an input unit that receives instructions from the user,
The teaching image generation apparatus according to claim 13, wherein the inspection area specifying unit specifies the inspection area based on the received user instruction.   The teaching image generation apparatus according to claim 13 or 14, wherein a position where the image is arranged is a position that does not overlap the inspection region image. The feature amount calculation unit calculates a gravity center position of the inspection region image,
The teaching position according to any one of claims 13 to 15, wherein the position calculation unit searches for and calculates a position where the image is to be arranged from the vicinity around the calculated gravity center position of the inspection region image. Image generation device. An inspection area determination unit that determines whether or not the inspection area is an area to be extracted;
The change inspection region image generation unit generates the change inspection region image in which the area of the inspection region image is reduced when it is determined that the inspection region is the region to be extracted. Item 17. The teaching image generation device according to any one of Items 16. An inspection area determination unit that determines whether or not the inspection area is an area to be extracted;
The change inspection region image generation unit generates the change inspection region image in which the area of the inspection region image is enlarged when it is determined that the inspection region is not the region to be extracted. The teaching image generation device according to any one of 16. An inspection area determination unit that determines whether or not the inspection area is an area to be extracted;
The feature amount calculation unit calculates a gravity center position of the inspection region image,
When it is determined that the inspection area is the area to be extracted, the image input unit newly inputs a new teaching image having the inspection area,
The image generation unit generates a new defect addition teaching image in which the change inspection region image is arranged at a position corresponding to the center of gravity of the inspection region image in the new teaching image. Item 16. The teaching image generation device according to any one of Items 15. A first inspection area determination unit that determines whether the inspection area is an area to be extracted;
The feature amount calculation unit calculates a gravity center position of the inspection region image,
When it is determined that the inspection area is the area to be extracted, the image input unit newly inputs a new teaching image having the inspection area,
A second inspection region determination unit that determines whether or not the inspection region of the new teaching image is the region to be extracted;
When it is determined that the inspection area of the new teaching image is not the area to be extracted, the image generation unit is located at a position corresponding to the center of gravity position of the inspection area image in the new teaching image. The teaching image generation apparatus according to any one of claims 13 to 15, which generates a new defect addition teaching image in which a change inspection area image is arranged. An area setting unit for setting first and second inspection surrounding areas around the inspection area image and the change inspection area image, respectively;
The first density value of the inspection area image, the second density value of the first background image in the background area other than the inspection area image, which is an area in the first inspection surrounding area, and in the second inspection surrounding area And a density value calculation unit that calculates a fourth density value of the change inspection area image based on a third density value of the second background image of the background area other than the change inspection area image. The teaching image generation apparatus according to any one of claims 13 to 20. A teaching image generation program for causing a computer to execute a process of generating a teaching image having an inspection area, in order to cause an algorithm that learns a self-developed target process in a teaching process.
Inputting a teaching image having the inspection region;
Designating the inspection area;
Calculating a feature amount of an inspection region image in the designated inspection region;
Calculating a position where the image is arranged in the teaching image;
Generating a changed inspection area image in which the feature amount of the inspection area image is changed;
A teaching image generation program causing a computer to execute a step of generating a new defect addition teaching image in which the change inspection area image is arranged at a position where the calculated image is arranged.   A recording medium on which the teaching image generation program according to claim 21 is recorded. Using a plurality of types of image processing filters prepared in advance, a new defect addition teaching image having one or more inspection areas generated by the teaching image generation method according to any one of claims 1 to 12 is previously stored. An image processing algorithm generation method for generating one or more final image processing algorithms for generating an image corresponding to a prepared target image,
Including a step of generating a plurality of image processing algorithms by repeating generations until a predetermined end condition is satisfied,
The image processing algorithm generated in the generation that satisfies the predetermined end condition is the one or more final image processing algorithms,
Each of the plurality of image processing algorithms is an algorithm that uses a plurality of selected image processing filters selected from the plurality of types of image processing filters without hindering duplication in a random order;
The step of generating the plurality of image processing algorithms of N (natural number) +1 generation includes:
Generating a plurality of N generations of processed images obtained by performing processing of the plurality of image processing algorithms of N generations on the new defect addition teaching image;
For each of the N generation image processing algorithms, calculating an evaluation value serving as an evaluation reference based on the target image and a corresponding N generation processed image;
Selecting one or more N generation image processing algorithms for which the evaluation value equal to or greater than a predetermined value is selected from among the plurality of N generation image processing algorithms;
Generating an image processing algorithm in which each of the selected one or more N generation image processing algorithms is changed so that the evaluation value is high as an N + 1 generation image processing algorithm. Algorithm generation method.   25. The image processing algorithm generation method according to claim 24, wherein the predetermined end condition is a condition in which the evaluation value of any one or more of the N generations of selected image processing algorithms is a predetermined value or more.   Among the plurality of image processing algorithms, one or more image processing algorithms other than the N generation selected one or more image processing algorithms are generated as an N + 1 generation one or more image processing algorithms using an evolutionary method. The image processing algorithm generation method according to claim 24, further comprising a step.   The image processing algorithm generation method according to claim 26, wherein the evolutionary method is a genetic algorithm.   The number of new defect addition teaching images generated by the teaching image generation method according to any one of claims 1 to 12 is greater than the number of inspection areas in the new defect addition teaching image. 27. The image processing algorithm generation method according to 27. A new defect addition teaching image having one or more inspection areas generated by the teaching image generating method according to any one of claims 1 to 11 using a plurality of types of image processing filters prepared in advance. An image processing algorithm generation device that generates one or more final image processing algorithms that generate an image corresponding to a prepared target image,
An image processing algorithm generation unit for generating a plurality of image processing algorithms;
The image processing algorithm generation unit repeats the process of generating the plurality of image processing algorithms while changing the generation until a predetermined end condition is satisfied,
The image processing algorithm generated in the generation that satisfies the predetermined end condition is the one or more final image processing algorithms,
Each of the plurality of image processing algorithms is an algorithm that uses a plurality of selected image processing filters selected from the plurality of types of image processing filters without hindering duplication in a random order;
The step of generating the plurality of image processing algorithms of N (natural number) +1 generation includes:
An image data conversion unit that generates a plurality of N generation post-processing images obtained by performing processing of the N generation of the plurality of image processing algorithms on the new defect addition teaching image;
For each of the N generations of image processing algorithms, an image processing algorithm performance evaluation unit that calculates an evaluation value serving as a reference for evaluation based on the target image and the corresponding N generations of processed images In addition,
The image processing algorithm performance evaluation unit selects one or more N generation image processing algorithms for which the evaluation value equal to or greater than a predetermined value is selected from the N generation image processing algorithms,
The image processing algorithm generation unit generates, as an N + 1 generation image processing algorithm, an image processing algorithm in which each of the selected one or more N generation image processing algorithms is changed so that the evaluation value increases. An image processing algorithm generation device.   30. The image processing algorithm generation device according to claim 29, wherein the predetermined end condition is a condition in which the evaluation value of any one or more of the N generations of selected image processing algorithms is equal to or greater than a predetermined value.   Among the plurality of image processing algorithms, one or more image processing algorithms other than the N generation selected one or more image processing algorithms are generated as an N + 1 generation one or more image processing algorithms using an evolutionary method. The image processing algorithm generation apparatus according to claim 29 or 30, further comprising an evolutionary image processing algorithm generation unit.   The number of new defect addition teaching images generated by the teaching image generation method according to any one of claims 1 to 12 is greater than the number of inspection areas in the new defect addition teaching image. 31. An image processing algorithm generation device according to 31. A target image prepared in advance for a new defect addition teaching image having one or more inspection areas generated by the teaching image generation program of claim 22 using a plurality of types of image processing filters prepared in advance. An image processing algorithm generation program for causing a computer to execute processing for generating one or more final image processing algorithms for generating an image corresponding to
Generating a plurality of image processing algorithms;
Repeating the step of generating the plurality of image processing algorithms while changing the generation until a predetermined end condition is satisfied,
The image processing algorithm generated in the generation that satisfies the predetermined end condition is the one or more final image processing algorithms,
Each of the plurality of image processing algorithms is an algorithm that uses a plurality of selected image processing filters selected from the plurality of types of image processing filters without hindering duplication in a random order;
The step of generating the plurality of image processing algorithms of N (natural number) +1 generation includes:
Generating a plurality of N generations of processed images obtained by performing processing of the plurality of image processing algorithms of N generations on the new defect addition teaching image;
For each of the N generation image processing algorithms, calculating an evaluation value serving as a reference for evaluation based on the target image and a corresponding N generation processed image;
Selecting one or more N generation image processing algorithms for which the evaluation value equal to or greater than a predetermined value is selected from among the plurality of N generation image processing algorithms;
Generating an image processing algorithm in which each of the selected one or more N generation image processing algorithms is changed so that the evaluation value becomes high as an N + 1 generation image processing algorithm. Algorithm generator.   A recording medium on which the image processing algorithm generation program according to claim 33 is recorded.   An image inspection method including a step of determining pass / fail of an image to be inspected using the image processing algorithm generated by the image processing algorithm generation method according to claim 24.   An image inspection apparatus including an image quality determination unit that determines quality of an inspection target image using the image processing algorithm generated by the image processing algorithm generation method according to claim 24.   An image inspection program for causing a computer to execute a step of determining pass / fail of an image to be inspected using the image processing algorithm generated by the image processing algorithm generation program according to claim 33.   A recording medium on which the image inspection program according to claim 37 is recorded.