JP2008015824A - Filter line generation device, filter line generation method, program for making computer function as filter line generation device, and storage medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To generate a superior filter line with minimized load in a short time. <P>SOLUTION: The filter line generation device 300 includes an input part 310 receiving input of original image data; a storage part 320; an arithmetic part 330 executing filter line generation processing; and an output part 350 outputting a filter line. The arithmetic part 330 includes a condition input part 331 receiving input of a processing termination condition; a filter line capability value calculation part 332 calculating a capability value showing the capability of image processing by each filter line; a filter line load value calculation part 333 calculating a load value showing the load on image processing to which each filter line is applied; a filter line evaluation value calculation part 334 calculating an evaluation value based on the capability value and the load value; a master filter line extraction part 335 extracting a filter line satisfying an extraction condition; a next-generation creation part 336; and a superior filter line output part 337. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to image processing, and more particularly, to generation of a filter row used in image processing. More specifically, the present invention relates to a filter string generation device and a filter string generation method for generating a filter string using weights based on the filter configuration, a program for causing a computer to function as the filter string generation device, and The present invention relates to a computer-readable recording medium storing a program.

  First, the actual development of the image processing algorithm will be described. In an inspection apparatus using an image or the like, when there is a request to extract a characteristic pattern in the obtained image, the developer creates a dedicated algorithm for extracting the pattern. The contents of the algorithm are a combination of filters that increase the feature amount of the target pattern, pattern extraction, and pattern information extraction.

  In developing a combination of filters, the developer selects an effective filter from existing image processing filters, determines the processing order, and the like, and the image is processed in a direction in which the feature amount of a desired pattern increases. Make adjustments as follows. Developers develop such image processing algorithms by trial and error.

  In the case of an inspection apparatus based on an image, if the object to be inspected changes, or if a new pattern to be extracted is added even if the object is the same, the person in charge sequentially analyzes the target pattern, It is necessary to develop an algorithm for extracting the pattern. However, this method has a problem that it is necessary for the person in charge of image processing to devise an algorithm every time the object is changed or added as described above, which requires a development period.

[Development of image processing algorithm using genetic algorithm]
As mentioned above, image processing algorithms require a great deal of man-hours for their development and introduction. As the image processing application field expands, there are several methods for generating image processing algorithms using optimization techniques. Proposed. The optimization method includes an evolutionary method that imitates the process of biological evolution, and an annealing method (simulated annealing).

  As a conventional technique for generating an image processing algorithm, there is an image processing algorithm generation method using a genetic algorithm (GA), which is one of the above evolutionary techniques. Hereinafter, the genetic algorithm will be described first.

  In the genetic algorithm, a certain combination of parameters to be optimized in a problem is expressed as one individual. An individual group formed by a plurality of individuals having different combinations of parameters forms one generation. Each individual has an evaluation value calculated using the evaluation function set in the problem. Generally, an individual closer to the solution to the problem has a higher evaluation value.

  Next, generating an N + 1th generation individual population by performing a genetic operation called wrinkle, crossover, or mutation on a certain Nth generation individual population is called generational change. In the process of evolution by repeating such generation changes, individuals with higher evaluation values have a higher probability of survival in the next generation. By sufficiently evolving an individual population, it is possible to generate an individual with a high evaluation value. The combination of parameters finally shown by the individual having the highest evaluation value is the optimal solution for this problem.

  Regarding a genetic algorithm, for example, Japanese Patent Laid-Open No. 9-178500 (Patent Document 1) discloses a car navigation device having a short route search time using a genetic algorithm. Japanese Patent Laying-Open No. 2004-258842 (Patent Document 2) discloses a parallel processing device using a genetic algorithm.

  Furthermore, regarding generation of an image processing algorithm using a genetic algorithm, for example, Japanese Patent Laid-Open No. 2004-38744 (Patent Document 3) describes a feature extraction method when learning data of a target image to be recognized is given. An image recognition algorithm generating apparatus that automatically generates and evaluates and supports development or generation of an algorithm for identifying a target image is disclosed. This apparatus includes an image processing program module database and an image processing program search unit. The image processing program search unit includes a recognition algorithm evaluation unit.

According to the image recognition algorithm generation device disclosed in Japanese Patent Application Laid-Open No. 2004-38744, image processing program modules are accumulated, and a recognition algorithm showing a high evaluation value is searched according to a search condition. Therefore, even if the developer of the image recognition algorithm does not know in detail the nature of the feature extraction method that varies depending on the procedure of the feature extraction process or the parameters of the feature extraction process, find an image recognition algorithm that achieves a high identification rate be able to.
JP-A-9-178500 JP 2004-258842 A JP 2004-38744 A

  In the technique as described above, the evaluation of the filter sequence to be obtained is determined by the content of the result image output using the filter sequence. In other words, although it is possible to search for a high-capacity filter string that outputs a desired processing result, evaluation is performed using a result image that is output using the filter string. May be generated.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a filter array generation apparatus that can generate an excellent filter array with a small load in a short time.

  Another object of the present invention is to provide a filter sequence generation device that generates a filter sequence for image processing in which results having similar quality are derived in a short time.

  Another object of the present invention is to provide a filter array generation method capable of generating an excellent filter array with a small load in a short time.

  Another object of the present invention is to provide a program for causing a computer to function as a filter array generation device that can generate an excellent filter array with a small load in a short time.

  Still another object of the present invention is to provide a recording medium storing a program for causing a computer to function as a filter array generation apparatus that can generate an excellent filter array with a small load in a short time.

  In order to solve the above-described problem, a filter array generation device according to an aspect of the present invention includes an ability value calculation unit that calculates an ability value representing an image processing ability for each of a plurality of filter arrays used for image processing. Prepare. Each filter row includes a plurality of filters. The filter sequence generation device includes, for each of the plurality of filter sequences, a load value calculation unit that calculates a load value representing a load on the image processing using the filter sequence, and a capability value and a load value for each of the plurality of filter sequences. Based on the evaluation value calculating means for calculating an evaluation value for evaluating the filter column, and extracting the filter column from each filter column based on each evaluation value calculated for each of the plurality of filter columns Extraction means, and generation means for generating a filter string of the next generation according to a genetic algorithm, with the filter string extracted by the extraction means as a parent generation.

  Preferably, the filter array generation device further includes an input unit that receives input of target image data prepared in advance as an image derived by performing image processing on the original image data and the original image data to be subjected to image processing. . The capability value calculating means calculates the capability value based on the result image data and the original image data obtained as a result of the image processing by the image processing means.

  Preferably, the capability value calculation means calculates a difference between the result image data and the original image data as the capability value.

  Preferably, the filter array generation device further includes a feature amount calculation unit that calculates a first feature amount based on the result image data and a second feature amount based on the original image data. The ability value calculation means calculates the ability value based on the first feature quantity and the second feature quantity.

  Preferably, the evaluation value calculation means includes calculation means for calculating the evaluation value by adding the capability value and the load value at a predetermined ratio.

  Preferably, the calculating means calculates a linear sum of the capacity value converted to the same dimension and the load value, a non-dimensionalizing means for converting the capacity value dimension and the load value dimension into the same dimension. Linear sum calculation means.

  Preferably, the load value calculation unit calculates the load value based on the number of filters included in the filter row.

  Preferably, the load value calculation unit calculates the load value based on a processing time required for image processing using the filter array.

  Preferably, the filter string generation device further includes an input unit that receives an input of a preset end condition, and an output unit that outputs a filter string having a maximum evaluation value when the end condition is satisfied.

  When the other situation of this invention is followed, the filter row | line | column production | generation method is provided. The filter sequence generation method includes a step of calculating a capability value representing the capability of image processing for each of a plurality of filter sequences used for image processing. Each filter row includes a plurality of filters. The filter sequence generation method calculates a load value representing a load on image processing using the filter sequence for each of the plurality of filter sequences, and based on the capability value and the load value for each of the plurality of filter sequences. A step of calculating an evaluation value for evaluating the filter column, a step of extracting the filter column from each filter column based on each evaluation value calculated for each of the plurality of filter columns, and an extraction step Generating the next generation filter string according to a genetic algorithm, with the extracted filter string as a parent generation.

  When the other situation of this invention is followed, the program for functioning a computer as a filter row | line | column production | generation apparatus is provided. This program causes a computer to execute a step of calculating a capability value representing the capability of image processing for each of a plurality of filter arrays used for image processing. Each filter row includes a plurality of filters. The program further includes: calculating a load value representing a load on image processing using the filter row for each of the plurality of filter rows; and a capability value and a load value for each of the plurality of filter rows. A step of calculating an evaluation value for evaluating the filter column, a step of extracting the filter column from each filter column based on each evaluation value calculated for each of the plurality of filter columns, and an extraction step And generating a next generation filter string according to a genetic algorithm, with the filter string extracted by the above as a parent generation.

  According to still another aspect of the present invention, a computer-readable recording medium storing the above program is provided.

  According to the filter array generation device of the present invention, an excellent filter array with a small load can be generated in a short time. In addition, according to the filter array generation device of the present invention, it is possible to generate a filter array for image processing in which results having similar quality are derived in a short time.

  According to the filter array generation method of the present invention, an excellent filter array with a small load can be generated in a short time. Furthermore, according to the program according to the present invention, it is possible to cause the computer to function as a filter row generation apparatus that can generate an excellent filter row with a small load in a short time. Further, when the recording medium according to the present invention is mounted on a computer and a program stored in the recording medium is executed, the computer operates as a device that generates an excellent filter array with a small load in a short time.

  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>
First, a combination of image processing filters in the image processing apparatus will be described with reference to FIG. FIG. 1 is a diagram conceptually showing a process for obtaining a result image 120 using image processing filters 110-1, 110-2,..., 110-n prepared in advance for the original image 100. The image processing filters 110-1 and 110-2 are image processing filters created in advance, for example, a smoothing filter, an intermediate value filter, a maximizing filter, a minimizing filter, a sobel filter, a Laplacian filter, and the like.

  When the filtering process is executed, the original image 100 is first given to the image processing apparatus. A target image corresponding to the original image 100 is also given in the same manner. Here, the target image is an image expected as an image obtained by processing the original image 100.

  As shown in FIG. 1, for example, when an n-th row of filters is given, first, filter processing using the first filter 110-1 is performed on the original image 100. As a result, filter processing using the filter 110-2 is performed on the image. Similarly, the process is repeated, and the process using the nth filter 110-n is executed. An image obtained by the last process is output as a result image 120. A feature amount is extracted from the result image 120, and an individual is evaluated based on the extracted feature amount. By repeatedly evolving generation changes, a filter row with a high recognition rate is finally output as an image processing algorithm.

  In this case, when the result image 120 is close to a preset target image, that is, when the result image 120 and the target image are quantified, and the difference between the values is smaller than a predetermined reference value, the result Each filter row used to obtain the image 120 is judged to be close to an algorithm for realizing image processing.

  With reference to FIG. 2, the case where the method using the genetic algorithm (GA) is applied to the image processing apparatus according to the present invention will be described. FIG. 2 is a diagram showing a mode of selecting a filter string to be used in the image processing apparatus by sequentially selecting a filter string composed of a plurality of filters according to a predetermined evaluation criterion.

  As shown in FIG. 2A, first-generation filter row groups 200-1, 200-2, etc. composed of a plurality of filter rows are first generated. Each filter row group is configured by combining filters prepared in advance at random. An original image 100 prepared in advance for each filter row of all the generated filter row groups is given as an input image. Each filter row group executes corresponding image processing by using each filter for the original image 100, and outputs result images 210-1, 210-2, and the like. Each result image is compared with a preset target image, and a filter row (for example, filter row 210-4) that outputs a result image close to the target image is given a high evaluation value as being close to the desired filter row. Here, as will be described later, the evaluation value is a value expressed in order to quantitatively express the attribute of each filter row as a result of the evaluation of the filter row. Specifically, the evaluation value is data that can be expressed quantitatively using, for example, an absolute sum of luminance differences, a sum of squares of the differences, or an average value, and the difference between the result image and the target image is The smaller the value, the larger the value. That is, the evaluation value is obtained by comparing a plurality of images obtained by applying the generated filter row to the original image with the target image. Further, “excellent” means a material having a large evaluation value. In the present embodiment, regarding the relationship between the evaluation value and the filter row, the performance of the filter row is improved when the evaluation value is increased, and the performance of the filter row is deteriorated when the evaluation value is reduced.

  A filter string given a high evaluation value in the first generation is regarded as a so-called excellent filter string in that generation and is registered as a parent for generating the next generation, that is, the second generation (filter string). 220).

  Next, as shown in FIG. 2B, the filter 220 that is an excellent filter row in the processing in the first generation is registered as the parent of each filter in the second generation. The second generation is generated from the parent by a technique such as crossover / mutation of GA. That is, a plurality of filter rows 220-1, 220-2, etc. are generated. Each filter row executes filter processing on the original image in the same manner as in the first generation, and outputs result images 230-1, 230-2, and the like. From each output result image, a filter row that outputs a result image closest to a preset target image is identified (for example, filter row 230-n) and registered as a parent in the next generation (filter row 240). .

  As shown in FIG. 2C, a filter sequence for realizing the image processing apparatus can be derived by repeating the same processing up to the Nth generation. In other words, the next generation filter row group is generated based on each filter constituting the filter row 240, and the filter row that outputs the result image closest to the target image is generated from the (N-1) generation filter row. As a result, filter row groups such as the Nth generation filter rows 250-1 and 250-2 are generated, and filter processing using the respective filter rows is executed. The filter row 260-2 that outputs the result image closest to the preset target image is derived as a filter row 270 for realizing the image processing apparatus.

  With reference to FIG. 3, the filter row generation device according to the first exemplary embodiment of the present invention will be described. FIG. 3 is a block diagram showing a functional configuration of filter array generation device 300. The filter string generation device 300 includes an input unit 310, a storage unit 320, a calculation unit 330, and an output unit 350.

  The input unit 310 accepts data input. The data includes original image data to be subjected to image processing. The storage unit 320 stores data input via the input unit 310, data generated by the arithmetic processing by the arithmetic unit 330, and data prepared in advance. The data prepared in advance includes data set for defining the operation of the filter array generation device 300 and a program for realizing each process described later. The recording unit 320 further stores a plurality of filters prepared in advance for image processing.

  The calculation unit 330 includes a condition input unit 331, a filter column capability value calculation unit 332, a filter column load value calculation unit 333, a filter column evaluation value calculation unit 334, a parent filter column extraction unit 335, and a next generation generation unit. 336 and an excellent filter row output unit 337. The filter string load value calculation unit 333 includes a filter number calculation unit 340.

  The condition input unit 331 stores the processing conditions input via the input unit 310 in an area secured in advance in the storage unit 320. In addition, the condition input unit 331 receives an input of the end condition of the process for generating the filter string, and applies the condition to the excellent filter string output unit 337. The filter row has a plurality of the aforementioned filters. The filter row can be generated within the range of filters prepared in advance by changing the filter to be combined.

  The filter column capability value calculation unit 332 calculates a capability value that represents the image processing capability of each filter column. The filter column load value calculation unit 333 calculates a load value representing a load on image processing to which each filter column is applied. For example, the filter number calculation unit 340 calculates the load value using the number of filters constituting the filter row based on the filter row stored in the storage unit 320. Specifically, the filter number calculation unit 340 calculates the reciprocal of the number of filters. Alternatively, the filter number calculation unit 340 calculates the load value by subtracting the number of filters from a preset constant.

  The filter column evaluation value calculation unit 334 evaluates the filter column based on the capability value calculated by the filter column capability value calculation unit 333 and the load value calculated by the filter column load value calculation unit 333. Is calculated. Preferably, in one aspect, filter row evaluation value calculation unit 334 calculates an evaluation value by adding the capability value and the load value at a predetermined ratio. The evaluation value is derived, for example, as a formula “evaluation value = a × capability value + b × load value” (a and b are constants).

  In another aspect, the magnitude relationship between the ability value and / or the load value and the magnitude relationship between the evaluation values may be inversely proportional. In this case, the filter string evaluation value calculation unit 334 calculates the evaluation value by adding the reciprocal of the capability value and / or the reciprocal of the load value. In this way, even if the magnitude relationship between the capability value and / or the load value and the magnitude relationship between the evaluation values are inversely proportional, the evaluation value is calculated appropriately.

  Specifically, the filter column evaluation value calculation unit 334 calculates the evaluation value without converting to the same dimension from the viewpoint of a so-called multi-objective optimization problem. In this case, it becomes an optimization problem formulated by a plurality of objective functions, and generally a perfect optimum solution cannot be obtained, but a plurality of solutions are obtained as so-called Pareto optimum solutions. The multi-objective optimization problem or the Pareto optimal solution is a well-known matter for those skilled in the art. Therefore, detailed description thereof will not be repeated here.

  Alternatively, the filter string evaluation value calculation unit 334 converts the dimension of the capability value and the dimension of the load value into the same dimension, and calculates a linear sum of the capability value converted into the same dimension and the load value.

  Based on the evaluation value calculated by the filter column evaluation value calculation unit 334, the parent filter column extraction unit 335 extracts a filter column satisfying a preset extraction condition from each filter column. Preferably, the parent filter row extraction unit 335 extracts a filter row having the maximum evaluation value.

  The next generation generation unit 336 generates the next generation filter sequence according to a known genetic algorithm, with the filter sequence extracted by the parent filter sequence extraction unit 335 as a parent generation. The excellent filter row output unit 337 outputs the filter row having the maximum evaluation value calculated for the filter row generated by the next generation generation unit 336. The excellent filter row output unit 337 preferably outputs a filter row having the maximum evaluation value when a preset end condition is satisfied.

  With reference to FIG. 4, the data structure of the filter string generation device 300 will be described. FIG. 4 is a diagram conceptually showing one mode of data storage in storage unit 320.

  Storage unit 320 includes an area 410 for storing a program and an area 420 for storing data. The area 410 further includes areas 411 to 417 for storing each program. The condition input program is stored in area 411. This program stores condition data input from the outside in an area secured in advance. This program associates with other programs whose condition data is referenced.

  The filter column capability value calculation program is stored in area 412. When this program is executed, the calculation unit 330 shown in FIG. 3 functions as the filter row capability value calculation unit 332.

  The filter column load value calculation program is stored in area 413. When this program is executed, the calculation unit 330 functions as the filter row load value calculation unit 333. The filter column evaluation value calculation program is stored in area 414. When this program is executed, the calculation unit 330 functions as the filter row evaluation value calculation unit 334. The parent filter string extraction program is stored in area 415. When this program is executed, the calculation unit 330 functions as the parent filter string extraction unit 335. The next generation generation program is stored in area 416. When this program is executed, the calculation unit 330 functions as the next generation generation unit 336. The excellent filter column output program is stored in area 417. When this program is executed, the calculation unit 330 functions as the excellent filter row output unit 337.

  Referring back to FIG. 4, the area 420 includes areas 421 to 424 that store data used for image processing. Data representing the filter column is stored in area 421. For example, N filter columns from filter columns 421-1 to 421-N are stored in the region 421.

  Filter data for image processing is stored in area 422. For example, n pieces of filter data 422-1 to 422-n are stored in the area 422.

  Image data is stored in area 423. Original image data to be subjected to image processing is stored in the area 423a. Data of a target image prepared in advance as an image derived by performing image processing on the original image is stored in the area 423b. The image data being processed is stored in area 423c. This data is data that is temporarily generated during the execution of image processing to which a filter row is applied, for example, until the processing for each filter is completed and the next processing is executed.

  Other data is stored in area 424. The other data includes an extraction condition set in advance for extracting a filter row, an end condition input in advance for defining the number of repetitions of image processing, and the like.

  Here, with reference to FIG. 5, a computer system 500 that operates as the filter array generation device 300 will be described. FIG. 5 is a block diagram showing a hardware configuration of computer system 500.

  The computer system 500 stores a CPU 510 connected to each other via a data bus, a mouse 520 and a keyboard 530 for receiving an instruction input from the outside, and data input from the outside or data temporarily generated by the CPU 510. A RAM 540, a hard disk 550 for storing data in a nonvolatile manner, a CD-ROM drive 570, a monitor 580, and a communication IF (interface) 590. A CD-ROM 572 is mounted on the CD-ROM drive 570. The computer system 500 is connected to the network 592 but need not be connected.

  The processing in the computer system 500 is realized by each hardware and software executed by the CPU 510. Such software may be stored in the hard disk 550 in advance. Alternatively, the software may be stored in a CD-ROM 572 or other recording medium and distributed as a program product. Furthermore, the software may be provided as a program product that can be downloaded by an information provider connected to the so-called Internet.

  Such software is temporarily stored in the hard disk 550 after being read from the recording medium by the CD-ROM drive 570 or other reading device or downloaded via the communication IF 590. The software is read from the hard disk 550 by the CPU 510 and stored in the RAM 540 in the form of an executable program. CPU 510 reads the program from RAM 540 and executes it.

  Each hardware constituting the computer system 500 shown in FIG. 5 is general. Therefore, the most essential part for realizing the calculation unit 330 of the filter array generation device 300 according to the present invention can be downloaded via the RAM 540, the hard disk 550, the software stored in the CD-ROM 572 or other recording medium, or the network. It can be said that it is a simple software.

  Since the operation of each hardware of the computer system 500 is well known, detailed description thereof will not be described.

  With reference to FIG. 6 and FIG. 7, the control structure of the filter array generation device will be described. FIG. 6 is a flowchart showing a procedure of basic processing executed by CPU 510 of computer system 500 that functions as filter row generation apparatus 300.

  In step S610, CPU 510 sets a condition for defining image processing based on externally input condition data. Specifically, CPU 510 stores data input via mouse 520, keyboard 530, FD drive device 560, CD-ROM drive device 570, or communication IF 590 in an area secured in advance in hard disk 550.

  In step S620, CPU 510 generates a plurality of filter rows of the initial generation by combining filters based on data stored in hard disk 550.

  In step S700, CPU 510 executes filter row group evaluation processing described later. When this process is executed, the evaluation value of each filter row is calculated.

  In step S630, CPU 510 determines whether or not a preset end condition is satisfied. If the end condition is satisfied (YES in step S630), the process ends. If not (NO in step S630), the process proceeds to step S640.

  In step S640, CPU 510 selects a parent filter column based on the previously generated evaluation value of the filter column group. This selection is performed, for example, by specifying a filter row having an evaluation value that is the maximum value from among the evaluation values of each filter row.

  In step S650, CPU 510 generates a next generation filter string group from the parent filter string using a genetic algorithm. Thereafter, the process returns to step S700.

  FIG. 7 is a flowchart showing a procedure of processing executed by CPU 510 for evaluating the filter column group.

  In step S710, CPU 510 calculates a capability value of each filter row included in the generated filter row group. Specifically, CPU 510 calculates a capability value based on original image data stored in hard disk 550 and result image data obtained as a result of image processing. For example, the CPU 550 calculates the difference between the result image data and the original image data. Or in another situation, CPU510 calculates the 1st feature-value based on result image data, and the 2nd feature-value based on original image data, respectively. The CPU 510 calculates a capability value using these feature amounts. The feature amount includes the feature amount of the feature region extracted from the target image data. This feature amount includes, for example, the position of the detection object to be detected from the image data, the contrast and size of the detection object, and the like.

  In step S720, CPU 510 calculates the load value of the filter row. Specifically, CPU 510 specifies the number of filters included in the filter row to be processed, and stores the number in an area reserved in advance in RAM 540.

  In step S730, CPU 510 calculates an evaluation value of the filter row. If the dimension of the ability value calculated in step S710 is different from the dimension of the load value calculated in step S720, CPU 510 performs an operation for aligning the dimension of the ability value and the dimension of the load value. For example, the ability value is expressed as the reciprocal of the number of pixels, or the value obtained by subtracting the number of pixels from the constant, and the load value is the reciprocal of the number of filters, or the number of filters is subtracted from the constant. When the value is expressed as a value obtained in this manner, the CPU 510 executes a process for normalizing the dimensions of the number of pixels and the number of filters using a constant identified in advance.

  In step S740, CPU 510 determines whether it is determined whether a preset end condition is satisfied. If the end condition is satisfied (YES in step S740), the process returns to the main process. If not (NO in step S740), the process returns to step S710.

  Here, with reference to FIG. 8, a process until the evaluation value of the filter row is calculated from the image data will be described. FIG. 8 is a diagram showing the flow of each process.

  First, a target image 810 and an original image 820 are prepared. The target image 810 shows an area to be extracted from the original image 820. The original image 820 is an image expressed by, for example, monochrome 8 bits, and has a gradation.

  When the filter row 830 is applied to the original image 820, a result image 850 is derived. The filter row 830 includes a plurality of filters as described above, and a plurality of filter rows 830 are generated by changing the combination of the filters. Image processing using each filter row is applied to the original image 820.

  The difference between the target image 810 and the result image 850 is calculated. Here, the difference 860 is a difference as an absolute value between the data of the target image 810 and the data of the result image 850. The difference 860 includes the number of pixels calculated as the difference in this case. A filter column capability value 870 is derived from this number of pixels. Specifically, the reciprocal of the number of pixels is calculated. Alternatively, the capacity value is calculated by subtracting the number of pixels from a preset constant. Alternatively, in another aspect, instead of using the number of pixels itself as the capability value, the difference pixel number / the number of pixels of the entire image may be used as the capability value.

  The evaluation value is calculated as a linear sum of the capacity value and the load value or a reciprocal of the capacity value and / or a reciprocal of the load value according to the magnitude relationship between the capacity value and / or the load value. .

  Thereafter, the integration process 880 of the filter string load value 840 and the filter string capability value 870 is performed, and a filter string evaluation value 890 is calculated.

  Next, the data structure of the filter array generation device 300 will be described with reference to FIGS. 9 and 10. FIG. 9 is a diagram illustrating an aspect of data storage in RAM 540 of computer system 500 that functions as filter row generation apparatus 300. The RAM 540 includes a table 900. The table 900 includes areas 910 to 950 for storing data.

  A filter column number for identifying the filter column is stored in area 910. For example, when N filter rows are generated, numerical values from “1” to “N” are stored in the area 910, respectively. Data for representing the contents of the filter included in the filter column is stored in area 920. For example, data for identifying a filter included in each filter column is stored in area 920 in accordance with the filter arrangement order.

  Data representing the ability value calculated when the image processing is executed is stored in area 930. The number of filters included in the filter row is stored in area 940. The evaluation value calculated as a result of the image processing described above is stored in area 950. When the processing for each filter column included in the filter column group is completed in this way, the data for each filter column is stored in each area, and a table 900 is generated. Each data is associated with a filter column number stored in the area 910. Therefore, when the filter column number is specified, the data stored in the areas 910 to 950 are specified.

  Here, calculation of the evaluation value will be described. The evaluation value is calculated as, for example, the formula “evaluation value = capability value + (10−number of filters) / 100”. In the case of the filter row having the filter row number “2”, the evaluation value “0.87” is obtained from the evaluation value = 0.85 + (10−2) / 100. Similarly, in the case of the filter row having the filter row number “7”, the evaluation value “0.89” is obtained from the evaluation value = 0.82 + (10−3) / 100.

  Note that the formula for calculating the evaluation value is not limited to the above, and a function identified in advance with respect to the relationship between the evaluation value and the ability value can be used. The function can be identified by, for example, an evaluation test of an inspection object.

  FIG. 10 is a diagram illustrating an aspect of data stored in RAM 540 when image processing is executed. The RAM 540 includes a table 1000. The table 1000 includes areas 1010 to 1050 for storing data. Each area is associated with each other.

  Data for identifying the filter column is stored in area 1010. Data for specifying a filter included in the filter column is stored in an area 1020 in accordance with the filter arrangement in the filter column. The capability value of the filter row is stored in area 1030. The number of filters included in the filter row is stored in area 1040. Here, the “ability value” is defined by a difference between an image obtained by processing an original image with a filter array and a target image. The smaller this difference, the higher the capacity of the filter array. For example, if the difference is “0”, the filter row is the best (or ideal) filter row.

  When the filter array generation device 300 starts image processing, data stored in the hard disk 550 is read by the CPU 510 and the data is temporarily stored in the RAM 540. Thereafter, image processing for each filter row is executed, and when the respective evaluation values are calculated, the evaluation values are sequentially stored in the area 1050.

  That is, as shown in FIG. 10B, the evaluation value of the image processing performed using the filter row is stored in the area 1050. Therefore, by referring to this evaluation value, a filter string that satisfies a predetermined evaluation condition is extracted.

  In this way, it is possible to generate a next-generation filter string by a genetic algorithm with a combination of filter strings that maximizes the evaluation value of the filter string in one generation as a parent. This makes it possible to search for an excellent filter row. In this case, a filter row having a higher effect can be generated by selecting, as a parent, a filter row having a small load from among filter rows having similar ability values.

  Therefore, with reference to FIG. 11, a method for generating a more effective filter row from filter rows having the same capability value will be described. FIG. 11 is a diagram conceptually showing a transition from a parent composed of a plurality of filter columns having the same capability value until a next generation filter column is generated.

  As shown in FIG. 11A, first, a filter row group is generated. From this filter string group, a filter string that is the parent of the genetic algorithm is selected. As described above, this filter row is a filter row having a small load, for example, among filter rows having close ability values. By selecting such a filter row as a parent, as shown in FIG. 11B, a more effective filter row is generated as a next-generation element, and an excellent filter row is generated in a short time. .

  As described above, according to the filter string generation device 300 according to the present embodiment, the evaluation value of the filter string is calculated from the capability value and the load value of the filter string. By generating the next generation using a genetic algorithm with the combination of the filter sequence that maximizes the evaluation value as a parent, the superior filter sequence is searched without inheriting the excellent filter combination without falling into a local solution. It becomes possible.

  Also, by selecting a filter row with a low load from among filter rows with similar ability values as a parent, a highly effective filter is generated as an element of the next generation, and an excellent filter row can be generated in a short time. . Furthermore, the same quality answer as before can be derived in a short time.

<Second Embodiment>
Hereinafter, a second embodiment of the present invention will be described. The filter sequence generation device according to the present embodiment replaces the value calculated based on the number of filters (the reciprocal of the number of filters, or the value obtained by subtracting the number of filters from a constant) with the processing time Is different from the filter array generation device 300 according to the first embodiment described above in that it has a function of calculating a load value that is a value obtained by subtracting the number of pixels from a constant. Note that the hardware configuration of the filter string generation device according to the present embodiment is the same as the hardware configuration of the filter string generation device 300 described above. Their functions are the same. Therefore, description thereof will not be repeated here.

  With reference to FIG. 12, the structure of the filter row | line | column production | generation apparatus 1200 which concerns on this Embodiment is demonstrated. FIG. 12 is a block diagram showing a functional configuration of filter array generation device 1200.

  The filter string generation device 1200 does not have the filter number calculation unit 340 in the configuration shown in FIG. The filter string generation device 1200 includes a processing time calculation unit 1240. The processing time calculation unit 1240 calculates the time required for processing using the filter array. This calculation is performed by adding a processing time derived in advance for each filter included in the filter row, as will be described later.

  With reference to FIG. 13 and FIG. 14, the data structure of filter sequence generation apparatus 1200 according to the present embodiment will be described. FIG. 13 is a diagram illustrating an aspect of data storage in the hard disk 550 of the computer system 500 that functions as the filter row generation apparatus 1200. Hard disk 550 includes areas 1310 and 1320 for storing data. These areas are related to each other.

  The filter number is stored in area 1310 as data for identifying the filter. The processing time required when image processing is executed by applying the filter to the image data is stored in area 1320. Therefore, the processing time of the filter is detected by specifying the filter by the filter number. The value calculated based on this processing time is used as the load value of the filter row.

  With reference to FIG. 14 and FIG. 15, the data structure of filter sequence generation apparatus 1200 according to the present embodiment will be described. FIG. 14 and FIG. 15 are diagrams showing one mode of data storage in RAM 540, respectively. The RAM 540 includes areas 1410 to 1450 for storing data.

  As shown in FIG. 14, data for identifying the filter column is stored in an area 1410. For example, the filter column number is used as this data. Data for specifying the content of the filter included in the filter column is stored in area 1420. This data is realized by data for identifying a filter included in the filter row, for example. The capability value of the filter column is stored in area 1430. The processing time required for the image processing to which the filter row is applied is stored in the area 1440. This processing time corresponds to the sum of the processing times of the filters included in the filter row. The evaluation value for the filter column is stored in area 1450. When the function as shown in FIG. 12 is realized, the ability value is calculated, and further the evaluation value is calculated. As a result, the values are stored in area 1430 and area 1450.

  That is, as shown in FIG. 15A, the capability value of each filter row is sequentially written in area 1430. Furthermore, an evaluation value is calculated based on the capability value in area 1430 and the processing time in area 1440. The evaluation value is calculated by, for example, the formula “evaluation value = ability value + (100−processing time) / 1000”. As shown in FIG. 15B, the evaluation values are stored in the area 1450, respectively.

  For example, the evaluation value “0.869” is obtained from the evaluation value = 0.85 + (100−81) / 1000 for the filter row with the filter row number “2”. Further, with respect to the filter row with the filter row number “7”, the evaluation value “0.885” is obtained from the evaluation value = 0.82 + (100−35) / 1000.

  As described above, according to the filter train generation device according to the present embodiment, a filter train that produces the maximum evaluation value of the filter train in one generation is generated. With this combination of filter rows as a parent, it is possible to generate the next generation by a genetic algorithm, and an excellent filter row search becomes possible. Furthermore, by selecting a filter row with a low load from among filter rows with similar ability values as a parent, a more effective filter can be generated as an element of the next generation, and an excellent filter row can be generated in a short time. it can. In addition, it is possible to derive an answer with the same quality as before in a short time.

  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.

It is a figure which represents notionally the process for obtaining a result image using the image process prepared with respect to the original image previously. It is a figure showing the aspect which selects the filter row | line for using for an image processing apparatus by selecting sequentially the filter row | line | column which consists of a some filter by the predetermined evaluation criteria. 3 is a block diagram illustrating a functional configuration of a filter string generation device 300. FIG. FIG. 3 is a diagram conceptually illustrating one aspect of data storage in a storage unit 320. 2 is a block diagram illustrating a hardware configuration of a computer system 500. FIG. 10 is a flowchart showing a procedure of basic processing executed by CPU 510 of computer system 500 functioning as filter row generation apparatus 300. It is a flowchart showing the procedure of the process which CPU510 performs in order to evaluate a filter row group. A process until the evaluation value of the filter row is calculated from the image data will be described. FIG. 10 is a diagram (part 1) showing one mode of data storage in RAM 540 of computer system 500 that functions as filter train generation device 300; FIG. 10 is a diagram (part 2) showing one mode of data storage in RAM 540 of computer system 500 functioning as filter train generation device 300; It is a figure which represents notionally the transition until the next generation filter row | line | column is produced | generated from the parent which consists of several filter row | line | columns with the same capability value. It is a block diagram showing the functional structure of the filter row | line | column production | generation apparatus 1200 which concerns on the 2nd Embodiment of this invention. It is a figure showing the one aspect | mode of the storage of the data in the hard disk 550 of the computer system 500 which functions as the filter row | line | column production | generation apparatus 1200. FIG. 11 is a diagram (part 1) showing one mode of data storage in RAM 540 of computer system 500 functioning as filter train generation device 1200; FIG. 11 is a diagram (part 2) showing one mode of data storage in RAM 540 of computer system 500 functioning as filter train generation device 1200;

Explanation of symbols

  300, 1200 Filter column generation device, 310 input unit, 320 storage unit, 330, 1230 calculation unit, 331 condition input unit, 332 filter column capability value calculation unit, 333 filter column load value calculation unit, 334 filter column evaluation value calculation unit 335 parent filter column extraction unit, 336 next generation generation unit, 337 excellent filter column output unit, 340 filter number calculation unit, 350 output unit, 500 computer system, 510 CPU, 520 mouse, 530 keyboard, 540 RAM, 550 hard disk, 570 CD-ROM drive, 572 CD-ROM, 580 monitor, 590 communication IF, 592 network, 1240 processing time calculation unit.

Claims (12)

For each of a plurality of filter columns used for image processing, it comprises capability value calculation means for calculating a capability value representing the capability of the image processing, each filter column including a plurality of filters,
Load value calculating means for calculating a load value representing a load on the image processing using the filter row for each of the plurality of filter rows;
Evaluation value calculating means for calculating an evaluation value for evaluating the filter string based on the capability value and the load value for each of the plurality of filter strings;
Extracting means for extracting a filter row from each of the filter rows based on the evaluation values calculated for each of the plurality of filter rows;
A filter sequence generation device comprising: generation means for generating a filter sequence of the next generation according to a genetic algorithm, with the filter sequence extracted by the extraction means as a parent generation. The image processing apparatus further includes input means for receiving input of target image data prepared in advance as original image data to be subjected to image processing and an image derived by performing image processing on the original image data,
The filter row generation device according to claim 1, wherein the capability value calculation unit calculates the capability value based on result image data obtained as a result of image processing by the image processing unit and the original image data.   The filter row generation device according to claim 2, wherein the capability value calculation means calculates a difference between the result image data and the original image data as the capability value. A feature amount calculating means for calculating a first feature amount based on the result image data and a second feature amount based on the original image data;
The filter row generation device according to claim 2, wherein the capability value calculation unit calculates the capability value based on the first feature amount and the second feature amount.   The filter string generation device according to claim 1, wherein the evaluation value calculation unit includes a calculation unit that calculates the evaluation value by adding the capability value and the load value at a predetermined ratio. The computing means is
Dimensionless means for converting the dimension of the capability value and the dimension of the load value into the same dimension;
The filter row generation device according to claim 5, further comprising: a linear sum calculation unit that calculates a linear sum of the capability value converted into the same dimension and a load value.   The filter sequence generation device according to claim 1, wherein the load value calculation unit calculates the load value based on the number of filters included in the filter sequence.   The filter row generation device according to claim 1, wherein the load value calculation unit calculates the load value based on a processing time required for image processing using the filter row. An input means for receiving an input of a preset end condition;
The filter string generation device according to claim 1, further comprising: an output unit that outputs a filter string having the maximum evaluation value when the termination condition is satisfied. For each of a plurality of filter columns used for image processing, the method includes a step of calculating a capability value representing the capability of the image processing, each filter column including a plurality of filters,
For each of the plurality of filter rows, calculating a load value representing a load on the image processing using the filter row;
For each of the plurality of filter rows, calculating an evaluation value for evaluating the filter row based on the capability value and the load value;
Extracting a filter row from each of the filter rows based on the evaluation values calculated for each of the plurality of filter rows;
And generating a next generation filter sequence according to a genetic algorithm, with the filter sequence extracted in the extraction step as a parent generation. A program for causing a computer to function as a filter row generation device, wherein the program causes the computer to
For each of a plurality of filter columns used for image processing, the step of calculating a capability value representing the capability of the image processing is executed, each filter column including a plurality of filters,
The program is further stored on the computer,
For each of the plurality of filter rows, calculating a load value representing a load on the image processing using the filter row;
For each of the plurality of filter rows, calculating an evaluation value for evaluating the filter row based on the capability value and the load value;
Extracting a filter row from each of the filter rows based on the evaluation values calculated for each of the plurality of filter rows;
A program for executing the step of generating a filter string of the next generation according to a genetic algorithm, with the filter string extracted in the extraction step as a parent generation.   A computer-readable recording medium storing the program according to claim 11.

Images