JP2012168641A - Image determination apparatus, image processor, camera and image determination program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image determination apparatus determining a criterion by learning.SOLUTION: An image determination apparatus comprises: a first storage part 16 memorizing a learning image; a feature extraction part 40 to acquire the concrete value of an indicator about image quality of the learning image; a display 14 displaying the learning image on a predetermined display condition; an input part 32 to acquire a determination result of an evaluator for the learning image displayed by the display; a second storage part 20 memorizing a learning element that is obtained by relating the determination result to the concrete value; and a learning operation part 36 determining a determination reference value of the evaluator about the indicator based on the learning element.

Description

  The present invention relates to an image determination device, an image processing device, a camera, and an image determination program.

  Digital cameras have a relatively low cost for recording media and can easily delete shot images without generating dust, so you can shoot as much as you want without having to worry about the increase in the number of shots. It has the advantage of being able to. For this reason, the number of images taken by a digital camera tends to increase as compared to a silver salt camera. On the other hand, as a result of the photographer taking pictures without fear of failure, the photographed images of the digital camera tend to include a relatively large number of failed images in which blurring or blurring has occurred.

  In order to prevent the failed image from being printed and the recording medium from being occupied by the failed image, the photographer selects the failed image on the camera LCD screen or PC monitor and sorts or deletes it. It can be carried out. However, performing such visual and manual sorting operations for many photographed images is cumbersome and burdens the photographer.

  As a conventional technique for solving such a problem, there is a technique for calculating an index for judging a photographing error by extracting a high-frequency component included in image data and automatically determining whether a photographed image is acceptable or not. It has been proposed (see Patent Document 1).

JP 2001-256498 A

  In the image determination technology according to the prior art, the determination criterion that is a criterion for sorting the captured images is only determined by, for example, statistically examining general trends of failed images and successful images. There is a problem that a deviation occurs with respect to the judgment standard required by the individual. This is because the criteria for sorting the photographed images vary depending on the individuality of the photographer, the purpose of the photographed image, the viewing size of the photographed image, and the like. In addition, even if the photographer can arbitrarily set the judgment criteria, since various factors affect the pass / fail judgment of the photographed image, the photographer himself recognizes what kind of criteria is suitable for himself, It is not easy to set correctly.

  The present invention has been made in view of such a situation, and an object thereof is to provide an image determination device or the like that determines a determination criterion by learning.

In order to achieve the above object, an image determination apparatus according to the present invention includes:
A first storage unit (16) for storing a learning image;
A feature extraction unit (40) for acquiring a specific value of an index relating to image quality for the learning image;
A display unit (14) for displaying the learning image under predetermined display conditions;
An input unit (32) for obtaining a judgment result of an evaluator for the learning image displayed by the display unit;
A second storage unit (20) for storing a learning element that associates the determination result with the specific value;
A learning calculation unit (36) for determining a judgment reference value of the evaluator regarding the index from the learning element.

For example, the feature extraction unit may include a principal component analysis unit (46).
The principal component analysis unit obtains provisional concrete values corresponding to two or more provisional indices relating to image quality for each of the two or more learning images, performs principal component analysis on the provisional concrete values, and performs two or more provisional values. You may calculate the said specific value regarding the main component with which the said temporary parameter | index was integrated.

  Further, for example, the learning calculation unit may perform discriminant analysis on the learning element and determine the determination reference value.

  For example, the image determination apparatus according to the present invention further includes an image size acquisition unit (34) that acquires the image viewing size information selected by the evaluator and reflects the image viewing size in the display condition. May be.

Further, for example, the image determination device according to the present invention may further include an image quality adjustment unit (38) for adjusting the image quality of the learning image,
The image quality adjustment unit adjusts the image quality of the learning image so that the specific value of the index of the learning image becomes a predetermined first value;
The display unit may display the learning image adjusted by the image quality adjustment unit.

An image processing apparatus according to the present invention includes any one of the image determination apparatuses described above,
A second specific value of a second index relating to image quality is acquired for the captured image, and the second specific value of the captured image is compared with the determination reference value determined by the image determination device. And a photographed image judging section (52) for judging the photographed image.

  A camera according to the present invention has the image processing apparatus described above.

In addition, an image determination program according to the present invention includes:
On the computer,
A procedure for reading the learning image stored in the first storage unit;
A procedure for calculating a specific value of an index related to image quality for the learning image;
Displaying the learning image under predetermined display conditions;
A procedure for obtaining a judgment result of an evaluator for the learning image displayed by the display unit;
Storing a learning element that associates the determination result with the specific value of the indicator;
And a procedure for determining a judgment reference value of the evaluator regarding the indicator from the learning element.

  In the above description, in order to explain the present invention in an easy-to-understand manner, the description is made in association with the reference numerals of the drawings showing the embodiments. However, the present invention is not limited to this. The configuration of the embodiment described later may be improved as appropriate, or at least a part of the configuration may be replaced with another component. Further, the configuration requirements that are not particularly limited with respect to the arrangement are not limited to the arrangement disclosed in the embodiment, and can be arranged at a position where the function can be achieved.

FIG. 1 is a block diagram of an image determination apparatus according to an embodiment of the present invention. FIG. 2 is a flowchart showing a first example of learning processing performed by the image determination apparatus shown in FIG. FIG. 3 is a conceptual diagram illustrating a determination reference value determination process in the learning process shown in FIG. FIG. 4 is a flowchart showing a second example of learning processing performed by the image determination apparatus shown in FIG. FIG. 5 is a conceptual diagram illustrating the learning process shown in FIG. FIG. 6 is a flowchart illustrating a third example of learning processing performed by the image determination apparatus shown in FIG. FIG. 7 is a conceptual diagram illustrating a procedure for changing the image quality of the learning image in the learning process shown in FIG. FIG. 8 is a block diagram of an image processing apparatus according to the second embodiment of the present invention. FIG. 9 is a conceptual diagram showing the relationship between learning processing and automatic discrimination processing performed in the image processing apparatus shown in FIG. FIG. 10 is a flowchart illustrating an example of automatic determination processing performed by the image processing apparatus illustrated in FIG.

First Embodiment FIG. 1 is a block diagram of an image determination apparatus 10 according to an embodiment of the present invention. The image determination apparatus 10 calculates a determination reference value used for an automatic determination process of a captured image by a learning process. The image determination device 10 may be mounted on a PC (personal computer) or the like, or may be mounted on an imaging device that generates a digital image, such as a digital camera or a mobile phone.

  The image determination device 10 includes an input device 12, a display device 14, a storage device 16, an image memory 18, a program memory 20, and a central processing device 30. The input device 12 is an operation unit that allows an evaluator using the image determination device 10 to transmit the evaluator's determination, will, and the like to the image determination device 10. Examples of the input device 12 include a mouse, a keyboard, a touch panel, and the like when the image determination device 10 is a PC, and a cross key, a dial, a button, and the like when the image determination device 10 is an imaging device. There is no particular limitation. The input device 12 outputs an operation signal to the central processing unit 30 based on the evaluator's operation.

  The display device 14 displays various images such as captured images and display images, operation menus, and the like. Examples of the display device 14 include a liquid crystal panel. The display device 14 is also controlled by the central processing unit 30 in the same manner as the input device 12. The storage device 16 stores various image files, a program for controlling the central processing unit 30 and the like, data used for the program, and the like. The image file stored in the storage device 16 includes a learning image used for learning processing by the image determination device 10, a captured image determined by the automatic determination processing by the captured image determination unit 52 shown in FIG.

  The program memory 20 stores a program for causing the image determination apparatus 10 to perform a learning process and a program for causing the image processing apparatus 50 illustrated in FIG. 7 to perform an automatic determination process. Further, the program memory 20 temporarily stores learning elements obtained in the process of calculating the final determination reference value during the learning process.

  The image memory 18 temporarily stores the image data stored in the storage device 16. The central processing unit 30 performs control so that the image data is stored in the image memory 18 when performing various arithmetic processes on the image data or when displaying an image on the display device 14.

  The central processing unit (CPU) 30 performs various arithmetic processes necessary for the learning process performed by the image determination apparatus 10. In the image determination device 10, the central processing unit 30 functions as a determination input processing unit 32, an image size acquisition unit 34, a learning calculation unit 36, an image quality adjustment unit 38, a feature extraction unit 40, and the like.

  The image size acquisition unit 34 acquires image viewing size information input by an evaluator who is a user of the image determination device 10. The image size acquisition unit 34 reflects the acquired image appreciation size information on display conditions (particularly display size) when the learning image is displayed on the display device 14. The image appreciation size information is information relating to under what conditions a captured image scheduled to be automatically determined using the learning process result is viewed.

  The image size acquisition unit 34 requests the evaluator to input image viewing size information via the input device 12. The image size acquisition unit 34 changes display conditions when the learning image is displayed on the display device 14 using the obtained image appreciation size information. For example, the image size acquisition unit 34 controls the display device 14 so that the learning image is displayed under a condition similar to or approximate to the use condition of the captured image processed by the automatic determination process performed thereafter.

  For example, if the image appreciation size information is information that the image is enlarged to a predetermined size and viewed, the image size acquisition unit 34 enlarges the display size of the learning image accordingly. On the other hand, if the image appreciation size information is information indicating that the image is to be reduced and viewed, the image size acquisition unit 34 reduces the display size of the learning image accordingly. By performing such a process, the determination reference value calculated by the learning process in the image determination apparatus 10 takes into consideration the image viewing size information and matches the reference required by the evaluator with high accuracy.

  The image quality adjustment unit 38 can adjust the image quality of the learning image. For example, the image quality adjustment unit 38 adjusts the image quality of the learning image by convolving a PSF (Point Spread Function) representing a blur / blur locus with respect to the original image that is the learning image before the image quality adjustment. . The image quality adjusted by the image quality adjustment unit 38 includes an edge amount (similar to that calculated by the edge amount calculation unit 42) and the like, but is not particularly limited.

As the blurred PSF used in the image quality adjustment unit 38, for example, a Gaussian distribution represented by the following equation (1-1) can be given. When image quality adjustment is performed using a Gaussian distribution, the image quality adjustment unit 38 normalizes the Gaussian distribution as represented by Expression (1-2). Further, the image quality adjustment unit 38 convolves the normalized Gaussian distribution with the luminance Y (i, j) of the learning image before the image quality adjustment. As a result, the image quality adjustment unit _obtains the luminance Y _gaussian (i, j) of the learning image after the image quality adjustment, as shown in Expression (1-3). Note that σ in Formula (1-1) can be calculated as a standard deviation and x = 4σ, but is not particularly limited. Moreover, T in Formula (1-3) represents transposition.

  In the learning process, when the image quality of the learning image is adjusted by the image quality adjustment unit 38, the display device 14 displays the adjusted learning image. By adjusting the image quality by the image quality adjustment unit 38, the image determination apparatus 10 can perform a learning process using a plurality of images that have the same composition but differ only in the index relating to the specific image quality. Is possible. Therefore, the image determination apparatus 10 can perform a learning process that eliminates or suppresses influences other than the index related to a specific image quality, and obtains a determination reference value that matches the reference required by the evaluator with high accuracy. Can do. When the image quality adjustment unit 38 performs adjustment so that the edge amount of the learning image becomes a predetermined value, the feature extraction unit 40 does not need to calculate the edge amount. This is because the feature extraction unit 40 can acquire the edge amount of the learning image only by reading the setting value of the image quality adjustment unit 38.

  The feature extraction unit 40 acquires a specific value of an index related to the image quality of the learning image used in the learning process. The feature extraction unit 40 includes an edge amount calculation unit 42, a data acquisition unit 44, and a principal component analysis unit 46. The edge amount calculation unit 42, the data acquisition unit 44, and the principal component analysis unit 46 acquire specific values of indices related to the image quality of the learning image by different methods.

  The index of image quality acquired by the feature extraction unit 40 is not particularly limited, and the feature extraction unit 40 is a specific index that is effective for determining whether the learning image is a success image or a failure image. A value can be obtained. Examples of indexes acquired by the feature extraction unit 40 include DCT coefficients, wavelet coefficients, and FFT coefficients in addition to the indexes acquired by the edge amount calculation unit 42, the data acquisition unit 44, and the principal component analysis unit 46 described later. For example, the real part.

  The edge amount calculation unit 42 calculates an edge amount, which is one of image quality indexes, from the learning image. An example of a method in which the edge amount calculation unit 42 acquires a specific value (specific value) of the edge amount from the learning image to be subjected to the learning process is expressed by Equations (2-1) to (2-4). Will be described below.

The edge amount calculation unit 42 calculates the luminance Y (i, j) of each pixel from the numerical values of the RGB channels in the learning image, as shown in Equation (2-1). Further, the edge amount calculation unit 42 convolves an edge detection filter with respect to the calculated luminance Y (i, j), as shown in Equations (2-2) and (2-3), and performs the x direction. And edges G _x (i, j) and G _y (i, j) in the y direction are calculated. The edge amount calculation unit 42 uses a Sobel filter as an edge detection filter. However, the edge detection filter used by the edge amount calculation unit 42 is not limited to this, and a Prewitt filter or the like can also be used. Furthermore, the edge amount calculation unit 42 calculates an edge amount G () that does not depend on the edge direction from the calculated edges G _x (i, j) and G _y (i, j) as shown in Equation (2-4). i, j) is calculated.

  The data acquisition unit 44 acquires, among the specific values of the image quality index related to the learning image, those recorded in the storage device 16 together with the learning image, the edge amount set in the image quality adjustment unit 38, and the like. Examples of the index acquired by the data acquisition unit 44 include a shutter speed and an EV value when a learning image is captured, an edge amount set in image quality adjustment in the image quality adjustment unit 38, and the like.

  The principal component analysis unit 46 performs principal component analysis on two or more indices (provisional indices) relating to the image quality of the learning image, and a principal component that is a new index that represents the overall tendency / feature of the learning image group to be subjected to learning processing. Is calculated. It is known that the principal component analysis unit 46 performs learning processing on two or more learning images in advance, and can be performed when specific values of two or more indices related to image quality can be acquired for each learning image. Hereinafter, a method by which the principal component analysis unit 46 calculates an index (principal component) relating to the image quality of the learning image will be described using Table 1 and Equations (3-1) to (3-3).

  As shown in Table 1, the principal component analysis unit 46 has specific values (provisional specific values) respectively corresponding to the index x, the index y, and the index z which are two or more (3 in the example shown in Table 1) related to image quality. Value) is acquired for each of two or more learning images (10 in the example shown in Table 1). The provisional index used by the principal component analysis unit 46 includes an edge amount, a shutter speed, an EV value, and the like, but is not particularly limited. As the provisional specific value, the specific value of the index calculated by the edge amount calculation unit 42 or the specific value of the index acquired by the data acquisition unit 44 can be used.

Next, the principal component analysis unit 46 obtains a covariance matrix V _xyz shown in Equation (3-1). In Equation (3-1), the variable a corresponds to the image number in Table 1, data represents the number of images, and when the example of Table 1 is applied to Equation (3-1), data is 10 Further, in Equation (3-1), specific values of the index x are shown in Table 1. Specific values of _{x a,} the index y Specific values of _{y a,} the index z is represented as _{z a.}

Further, the principal component analysis unit 46 obtains eigenvalues λ _α , λ _β , λ _γ of the covariance matrix V _xyz and eigenvectors u _α , u _β , u _γ corresponding to these eigenvalues. Further, the principal component analysis unit 46 calculates the calculated eigenvectors u _α , u _β , u _γ , the specific values x _a , y _a , z _a of the provisional indices for each image and the average value for each provisional index of the image group. From these, the specific values ξ _α , ψ _α , ζ _α of the principal components of each image are calculated. Eigenvector u _alpha, u _beta, the components of u _gamma, When _{u α 1~u α 3, u β} 1~u β 3, u γ 1~u γ 3 as shown in Equation (3-2), each The specific values ξ _α , ψ _α , ζ _α of the main components of the image can be expressed by Equation (3-3).

The specific values ξ _α , ψ _α , and ζ _α of each principal component calculated by the principal component analysis unit 46 are specific values of other indexes acquired by the feature extraction unit 40 as specific values of indexes related to the image quality of the learning image. Like the value, it can be used in the learning process. In addition, the principal component analysis unit 46 can calculate the contribution rate of each principal component from the eigenvalues λ _α , λ _β , and λ _γ, and can adopt only the principal component having a high contribution rate as an index used in the learning process. In this case, the principal component analysis unit 46 may adopt a principal component or a combination of principal components having a cumulative contribution rate exceeding 80% as an index used for the learning process. Of the principal components calculated by the principal component analysis unit 46, only a part of the principal components having a high contribution rate are used for the learning process, so that the judgment reference value is calculated by the learning calculation unit 36 or after the learning process. The calculation processing of the automatic image determination processing that is performed can be speeded up.

  The determination input processing unit 32 acquires the evaluator's determination result for the learning image displayed by the display device 14. For example, the determination input processing unit 32 requests the evaluator to determine whether the learning image displayed by the display device 14 is a success image or a failure image (that is, OK or NG). The evaluation result of the evaluator for the learning image is input to the determination input processing unit 32 via the input device 12.

  Further, the determination input processing unit 32 associates the obtained evaluation result of the evaluator with the specific value of the index relating to the image quality extracted by the feature extraction unit 40 for the learning image. Further, the determination input processing unit 32 stores in the program memory 20 as a learning element a correspondence between the determination result of the learning image and the specific value of the index. In this case, the type of index stored as the learning element may be one type or two or more types.

  The learning calculation unit 36 determines an evaluator's criterion value for the index corresponding to the specific value of the learning element from the learning element stored in the program memory 20. For example, when the index employed in the learning process is the edge amount, the learning element stored in the program memory 20 determines that the image of the edge amount is OK by the evaluator, and the image of any edge amount is NG. This is a specific record of the judgment. The learning calculation unit 36 obtains a value based on the assumption that an evaluator's OK / NG determination is separated from such a learning element using a certain edge amount value as a boundary, and uses the value as a determination reference value.

  The learning calculation unit 36 can determine the determination reference value by, for example, linear discriminant, but the determination reference value determination method is not particularly limited. Note that the determination reference value, the specific value, and the provisional specific value in the learning process include not only a scalar amount configured by one value but also an array configured by a plurality of values.

  FIG. 2 is a flowchart showing a first example of learning processing performed by the image determination apparatus 10 shown in FIG. In step 001, learning processing by the image determination apparatus 10 is started.

  In step 002, the image size acquisition unit 34 shown in FIG. 1 acquires viewing size information. The image size acquisition unit 34 prompts the evaluator to input viewing size information by a method such as displaying a selection menu on the display device 14. The evaluator inputs viewing size information such as a print size (L version, postcard size, A4, etc.) and a display size (100%, 50%, for a blog) via the input device 12. At this time, the image size acquisition unit 34 may store the acquired viewing size information in the program memory 20.

  In step S003, the central processing unit 30 shown in FIG. 1 determines a learning image. The central processing unit 30 prompts the evaluator to specify a learning image to be used in the learning process by a method such as displaying a comment on the display device 14. The evaluator designates an image prepared in advance in the storage device 16 or a captured image taken by the evaluator himself and stored in the storage device 16 as a learning image. The central processing unit 30 specifies the learning image designated by the evaluator as a learning image used for the learning process. Further, the central processing unit 30 temporarily stores the determined learning image in the image memory 18.

  At this time, the evaluator may specify one learning image or a plurality of learning images. In addition, a sample image suitable for the learning process may be stored in the storage device 16 in advance, and the evaluator may select a learning image from such sample images. Furthermore, the sample image stored in the storage device 16 may be composed of a plurality of images with different shooting scenes (night view, macro, sport, etc.). As a result, the evaluator can specify an image of the same shooting scene as the captured image to be automatically determined as a learning image. In addition, it is preferable that the image quality of a sample image has moderate dispersion | variation.

  In step S004, the feature extraction unit 40 acquires the specific value of the index related to the image quality for the learning image determined in step S003. There are various indexes used in the learning process. In the first example shown in FIG. 2, the case where the edge amount and the shutter speed are used as an index relating to image quality will be described as an example. In step S004, the edge amount calculation unit 42 of the feature extraction unit 40 calculates the edge amount of the learning image. Further, in step S004, the data acquisition unit 44 searches for information stored in the storage device 16 together with the learning image, and acquires the shutter speed of the learning image. The specific value of the edge amount calculated by the edge amount calculation unit 42 and the specific value of the shutter speed acquired by the data acquisition unit 44 may be temporarily stored in the program memory 20.

  In step S005, the central processing unit 30 controls the display device 14 to display the learning image determined in step S003. At this time, the central processing unit 30 designates the display condition of the learning image based on the viewing size information acquired in step S002. For example, if the image viewing size information acquired in step S002 is information indicating that the image is to be magnified to 200%, the central processing unit 30 causes the display device 14 to display the learning image at 200%. Specify display conditions. Thereby, the display device 14 displays the learning image under display conditions based on the viewing size information.

  In step S006, the determination input processing unit 32 of the central processing unit 30 acquires the evaluator's determination result for the learning image displayed by the display device 14. The determination input processing unit 32 requests the evaluator to input whether the learning image displayed by the display device 14 is OK or NG, and obtains the input information, thereby obtaining the evaluation result of the evaluator. get.

  In step S007, the determination input processing unit 32 associates the determination result of the evaluator obtained in step S006 with the specific values of the edge amount and shutter speed of the learning image acquired in step S004, and generates a learning element. . The determination input processing unit 32 also stores a learning element that associates the determination result of the learning image with the edge amount and the shutter speed in the program memory 20.

  In step S008, the central processing unit 30 requests the evaluator to input whether to end the learning process. When the central processing unit 30 has acquired input information for ending the learning process, the central processing unit 30 performs the process of step S009. On the other hand, when the input information indicating that the learning process is to be continued is acquired, the central processing unit 30 returns to the process of step S003, changes the learning image, and repeats the processes of steps S003 to S007.

  In step S <b> 009, the learning calculation unit 36 of the central processing unit 30 determines the evaluator's determination reference value related to the edge amount and the shutter speed from the learning elements stored in the program memory 20. FIG. 3 conceptually shows processing performed in determination of the determination reference value performed in the learning calculation unit 36. The graph shown in FIG. 3 is a plot of learning elements with the horizontal axis as the edge amount (index A) and the vertical axis as the shutter speed (index B). Black circles and white circles shown in FIG. 3 correspond to learning elements stored in the program memory 20 in step S007.

  In the graph shown in FIG. 3, eleven learning elements are plotted, and this is determined by determining learning images (step S003) to storing learning elements (step S007) for eleven learning images by step S009. This means that the process up to has been performed. In the graph shown in FIG. 3, whether it is a white circle or a black circle corresponds to the determination result of the learning image in the learning element. That is, the white circle is a learning element that the evaluator determines to be OK in step S006, and the black circle is a learning element that is determined to be NG. Further, the position of the horizontal axis of each plot corresponds to the specific value of the edge amount (index A) of the learning element, and the position of the vertical axis of each plot represents the specific value of the shutter speed (index B) of the learning element. It corresponds to.

  The learning calculation unit 36 determines that the evaluator determines that the image in the lower right region from the straight line 70 is OK and determines that the image in the upper left region from the straight line 70 is NG by the linear discrimination. Further, the learning calculation unit 36 determines that a criterion value that is a boundary for classifying the image into OK or NG with respect to the edge amount (index A) and the shutter speed (index B) by the evaluator is a straight line 70. In the case shown in FIG. 3, the criterion value calculated in step S009 is expressed in the form of a linear expression. The criterion value calculated in step S009 is stored in the storage device 16.

  In step S010, the central processing unit 30 ends a series of learning processes.

  As described above, the image determination apparatus 10 illustrated in FIG. 1 can learn and recognize the determination tendency of the evaluator who is a specific individual by performing a series of learning processes as illustrated in FIG. Therefore, the image determination apparatus 10 can determine a determination reference value that matches with high accuracy the determination reference required by the individual evaluator without requiring any expert knowledge or the like.

  Further, the image determination apparatus 10 reflects the image viewing size information acquired by the image size acquisition unit 34 in the display conditions for displaying the learning image. Accordingly, the image determination apparatus 10 can perform the learning process in consideration of the use of the captured image that is the target of the automatic determination process performed using the learning result. Therefore, the image determination apparatus 10 can determine a determination reference value according to the purpose.

  FIG. 4 is a flowchart illustrating a second example of learning processing performed by the image determination apparatus 10 illustrated in FIG. Steps S101 and S102 in FIG. 4 are the same as steps S001 and S002 described in FIG.

  In step S103 shown in FIG. 4, the central processing unit 30 shown in FIG. 1 acquires a learning image group composed of all the learning images used for determination criterion value determination (S110). The central processing unit 30 prompts the evaluator to specify all learning images used in the learning process by a method such as displaying a comment on the display device 14. The evaluator designates the learning image group by designating a directory in which the learning image group is stored. The central processing unit 30 specifies the learning image group designated by the evaluator as the learning image group used for the learning process.

  In step S104, the central processing unit 30 shown in FIG. 1 determines a learning image. The central processing unit 30 can automatically select one learning image from the learning images included in the learning image group acquired in step S103. However, the central processing unit 30 may have the evaluator select one learning image from the learning images included in the learning image group and use it as a learning image.

  In step S105, for each of the learning images included in the learning image group acquired in step S103, the edge amount calculation unit 42 in the feature extraction unit 40 calculates the edge amount, and the data acquisition unit 44 determines the shutter speed of the learning image. get. The edge amount calculation processing by the edge amount calculation unit 42 and the shutter speed acquisition processing by the data acquisition unit 44 are the same as the processing described in step S004 shown in FIG.

  In step S <b> 105, the principal component analysis unit 46 acquires the edge amount and shutter speed acquired by the edge amount calculation unit 42 and the data acquisition unit 44. Furthermore, the principal component analysis unit 46 performs principal component analysis on specific values of the edge amount and shutter speed of each learning image included in the learning image group, and calculates a principal component in which the edge amount and shutter speed are integrated. Then, in step S105, the specific value of the principal component is calculated for the learning image determined in step S104.

  FIG. 5 conceptually shows processing from principal component analysis (step S105) to determination of a criterion value (step S110) in the learning processing shown in FIG. In step S <b> 105, the principal component analysis unit 46 calculates the first principal component represented by the straight line 72. The first principal component is an index in which the edge amount (feature A) and the shutter speed (feature B), which are provisional indices in the learning process shown in FIG. 4, are integrated (see Formula 3-3). In the learning process shown in FIG. 4, since the contribution ratio of the first principal component represented by the straight line 72 is high, only the first principal component is used as an index.

  For the learning image determined in step S104, the specific value of the first principal component calculated in step S105 is represented as a point (scalar amount) arranged on the straight line 72. A calculation formula for calculating the specific value of the first principal component is stored in the program memory 20.

  The processes related to steps S106 to S108 are the same as steps S005 to S007 described in FIG.

  In step S109, central processing unit 30 determines whether or not learning elements have been stored for all learning images included in the learning image group selected in step S103. When the central processing unit 30 recognizes that learning elements are stored for all learning images, the central processing unit 30 performs the process of step S110. On the other hand, when it is recognized that the learning image in which the learning element is not stored remains, the central processing unit 30 changes the learning image and repeats the processes of steps S104 to S108. However, the central processing unit 30 may acquire the will of the evaluator via the input device 12, and may proceed to step S110 regardless of whether or not there remains a learning image in which no learning element is stored.

  In the second and subsequent steps S105, since the first principal component has already been calculated in the first loop, there is no need to perform principal component analysis again, and the first main image is selected for the newly selected learning image. It is sufficient to calculate the specific values of the components.

  In step S <b> 110, the learning calculation unit 36 of the central processing unit 30 determines the evaluator's determination reference value related to the first principal component from the learning elements stored in the program memory 20. As shown in FIG. 5, the learning element acquired in the learning process shown in FIG. 4 is a correspondence between the specific value of the first principal component and the evaluator's determination result. Note that the meanings of the plots (white circles and black circles) in FIG. 5 are the same as those in FIG.

  The learning calculation unit 36 recognizes that the image in the lower right part of the straight line 72 is OK and the image in the upper left area of the straight line 72 has a tendency to be determined as NG by linear discrimination. Further, the learning calculation unit 36 determines that the criterion value that is a boundary for classifying the image into OK or NG for the first principal component is the point 74. In the case illustrated in FIG. 5, the determination reference value calculated in step S <b> 110 is represented by a single numerical value. The criterion value calculated in step S110 is stored in the storage device 16.

  In step S111, the central processing unit 30 ends a series of learning processes.

  As described above, the image determination apparatus 10 illustrated in FIG. 1 can speed up the calculation process of the determination reference value in the learning calculation unit 36 by performing the learning process using the principal component analysis.

  FIG. 6 is a flowchart illustrating a third example of learning processing performed by the image determination apparatus 10 illustrated in FIG. Steps S201 and S202 in FIG. 6 are the same as steps S001 and S002 described in FIG.

  In step S203, the central processing unit 30 shown in FIG. 1 determines a learning image (original image). The central processing unit 30 asks the evaluator to select a learning image (original image) specified for the learning process from the sample images prepared in advance, and the image selected by the evaluator is used as the learning image. Specified as (original image).

  At this time, the evaluator designates one learning image (original image). The sample image prepared in advance as a learning image (original image) preferably has a specific value of an image quality index (for example, an edge amount) after image quality adjustment performed in step S204. . Note that the learning image (original image) is temporarily stored in the image memory 18.

  In step S204, the image quality adjustment unit 38 of the central processing unit 30 shown in FIG. 1 adjusts the image quality of the learning image (original image) specified in step S204 to generate a learning image (after image quality adjustment). The image quality adjustment unit 38 adjusts the image quality of the learning image (original image) by convolving a PSF (Point Spread Function) representing a blur / blur locus, and generates a learning image (after image quality adjustment).

  FIG. 7 is a conceptual diagram illustrating a procedure for changing the image quality of the learning image in the learning process shown in FIG. The learning image (original image) selected in step S203 is a general judgment in which, in the first step S204, an index of the image quality (edge amount in the example shown in FIG. 7) is statistically calculated. Adjustment is made so that the value (point D) closest to the reference value (point 76) is obtained. That is, the image quality adjustment unit 38 sets the specific value of the edge amount of the learning image (after image quality adjustment) generated by the image quality adjustment to the point D, and performs image quality adjustment. Note that the edge amount of the learning image (after image quality adjustment) (setting value of image quality adjustment performed in step S204) is stored in the program memory 20.

  In step S205, the data acquisition unit 44 of the feature extraction unit 40 acquires the edge amount of the learning image (after image quality adjustment) stored in the program memory 20. In the third example illustrated in FIG. 6, the case where only the edge amount is used as an index relating to image quality will be described as an example.

  The processes related to steps S206 to S208 shown in FIG. 6 are the same as steps S005 to S007 described in FIG.

  In step S209, the central processing unit 30 examines the learning elements stored in the program memory 20, and determines whether or not the determination reference value may be determined based on the already calculated learning elements. For example, when the learning element stored in the program memory 20 includes both an OK evaluation and an NG determination by the evaluator, the central processing unit 30 uses the already calculated learning element. It is determined that the determination reference value may be determined. On the other hand, the central processing unit 30 determines the determination reference value when the learning element stored in the program memory 20 does not include one of the OK and NG determinations by the evaluator. Judge that it should not.

  If the central processing unit 30 determines in step S209 that the determination reference value may be determined, the process proceeds to step S211. On the contrary, if the central processing unit 39 determines that the determination reference value should not be determined, the process proceeds to step S210.

  In step S210, the image quality adjustment unit 38 of the central processing unit 30 changes the setting value of the image quality adjustment performed in step S204. In case 1 of FIG. 7, the setting value of the first image quality adjustment (step S204) is point D, and the evaluator's judgment acquired in step S207 for the learning image (after image quality adjustment) was OK. Is the case. In such a case, in step S210, the setting value of the second image quality adjustment (step S204) is changed to a point C that is worse (smaller edge amount) than the first setting value. Case 2 in FIG. 7 is a case where the first evaluator's judgment is NG, contrary to case 1, and the setting value of the second image quality adjustment (step S204) is the first setting. Change to a point E that is better than the value (more edge amount).

  After the image quality adjustment setting value is changed in step S210, the processing in steps S204 to S208 is repeated as shown in FIG.

  In step S211, the central processing unit 30 requests the evaluator to input whether to end the learning process. When the central processing unit 30 has acquired input information for ending the learning process, the central processing unit 30 performs the process of step S212. On the other hand, when the input information indicating that the learning process is to be continued is acquired, the central processing unit 30 returns to the process of step S203, changes the learning image (original image), and repeats the processes of step S203 to step S210.

  In step S <b> 212, the learning calculation unit 36 of the central processing unit 30 determines the evaluator's judgment reference value related to the edge amount from the learning elements stored in the program memory 20. In case 1 of FIG. 7, the edge amount of the learning image that is finally determined to be OK is the point D, and the edge amount of the learning image that is determined to be NG for the first time is the point C. In this case, the learning calculation unit 36 sets a point P that is an intermediate point between the point C and the point D as a determination reference value. In case 2 of FIG. 7, the edge amount of the learning image that is finally determined to be NG is the point E, and the edge amount of the learning image that is determined to be OK for the first time is the point F. In this case, the learning calculation unit 36 sets a point Q, which is an intermediate point between the points E and F, as a determination reference value. The criterion value calculated in step S212 is stored in the storage device 16.

  In step S213, the central processing unit 30 ends a series of learning processes.

  As described above, the image determination apparatus 10 illustrated in FIG. 1 can perform the learning process by adjusting the image quality by the image quality adjustment unit 38 and using learning images that have the same composition but differ only in the image quality. it can. Therefore, the image determination device 10 can suppress influences other than the index relating to the specific image quality, and can obtain a determination reference value that matches the reference required by the evaluator with high accuracy.

  In addition, the image determination apparatus 10 can appropriately distribute the specific values of the learning image index by performing image quality adjustment by the image quality adjustment unit 38. Thereby, the image determination apparatus 10 can reduce the time and labor required for the evaluator in the learning process. Further, the image determination apparatus 10 evaluates the edge amount of the learning image (after the image quality adjustment) obtained by the first image quality adjustment by making the edge amount close to a statistically calculated general determination reference value. It is possible to further reduce the time and labor required for the learning process.

Second Embodiment FIG. 8 is a block diagram of an image processing apparatus 50 according to a second embodiment of the present invention. The image processing apparatus 50 is different from the image determination apparatus 10 shown in FIG. 2 in that the central processing unit 30a can perform not only the calculation process related to the learning process but also the calculation process related to the automatic determination process. 10 is the same.

  As shown in FIG. 8, the central processing unit 30a of the image processing apparatus 50 can function not only as a processing unit similar to the central processing unit 30 shown in FIG. 1, but also as a captured image determination unit 52. . The captured image determination unit 52 includes a second feature extraction unit 60 and a comparison unit 54.

  FIG. 9 is a conceptual diagram showing the relationship between learning processing and automatic discrimination processing performed in the image processing apparatus 50 shown in FIG. The captured image determination unit 52 performs automatic determination processing of a captured image using the determination reference value that is the evaluation tendency of the evaluator calculated in the learning process described in the first embodiment. Further, the central processing unit 30a of the image processing apparatus 50 can perform necessary processing such as classification of captured images based on the result of automatic discrimination processing by the captured image determination unit 52.

  The second feature extraction unit 60 illustrated in FIG. 8 can acquire a specific value (second specific value) of an index (second index) related to image quality for a captured image that is a target of automatic discrimination processing. The second feature extraction unit 60 includes a second edge amount calculation unit 62 and a second data acquisition unit 64. The second edge amount calculation unit 62 and the second data acquisition unit 64 are the edge amount calculation unit 42 and the data acquisition except that the target for acquiring the index is not a learning image but a captured image that is a target of automatic discrimination processing. This is the same as the unit 44.

  The comparison unit 54 compares the specific value (second specific value) of the index (second index) of the captured image that is the target of the automatic determination process with the determination reference value determined by the learning calculation unit 36 through the learning process. Then, it is determined whether the captured image is a success image or a failure image (ie, OK or NG).

  FIG. 10 is a flowchart showing an example of the automatic determination process shown in FIG. In step S301 shown in FIG. 1, the automatic discrimination process by the image processing apparatus 50 is started.

  In step S <b> 302, the central processing unit 30 a acquires a determination reference value used for automatic determination processing performed by the captured image determination unit 52. In step S302, the learning process described in FIGS. 2, 4, and 6 may be performed, and the determination reference value may be acquired by reading the determination reference value stored in advance in the storage device 16 or the like. .

  In step S302, the central processing unit 30a determines a captured image to be subjected to automatic discrimination processing. The central processing unit 30a urges the evaluator to designate a captured image that is the target of the automatic discrimination process by a method such as displaying a comment on the display device 14. The evaluator designates a captured image already stored in the storage device 16 or a captured image newly stored in the storage device 16. The central processing unit 30a specifies the captured image designated by the evaluator as the captured image that is the target of the automatic discrimination process. Further, the central processing unit 30a temporarily stores the determined captured image in the image memory 18. At this time, the evaluator may specify one captured image or a plurality of captured images.

  In step S304, the second feature extraction unit 60 in the photographed image determination unit 52 acquires the specific value (second concrete value) of the index (second index) related to the image quality for the photographed image determined in step S303. The specific value (second specific value) acquired in step S304 is the same as the index used in the learning process when determining the determination reference value so that it can be compared with the determination reference value acquired in step S302. A specific value relating to the index is preferable. For example, if the index used in the learning process is the edge amount and the shutter speed, the second edge amount calculation unit 62 of the second feature extraction unit 60 calculates the edge amount of the captured image, and the second acquisition unit 64 captures the image. Get the shutter speed of the image.

  In step S305, the comparison unit 54 in the captured image determination unit 52 compares the specific value (second specific value) of the captured image acquired in step S304 with the determination reference value acquired in step S302. Accordingly, the comparison unit 54 determines whether the captured image that is the target of the automatic determination process is OK or NG. In step S305, if the comparison unit 54 determines that the captured image is OK, the process proceeds to step S306, and if it is determined that the captured image is NG, the process proceeds to step S307.

  In steps S306 and S307, the central processing unit 30a classifies the captured images based on the result of the automatic discrimination process. That is, in step S306, the central processing unit 30a classifies the captured images determined to be OK into an OK image folder created in the storage device 16. In step S307, the central processing unit 30a classifies the captured images determined to be NG into NG image folders. In step S307, the central processing unit 30a may perform processing for deleting the captured image determined to be NG from the storage device 16.

  In step S308, the central processing unit 30a requests the evaluator to input whether or not to end the automatic discrimination process. When the central processing unit 30a has acquired input information for ending the automatic discrimination process, the central processing unit 30a proceeds to the process of step S309. On the other hand, when the input information indicating that the automatic discrimination processing is continued is acquired, the central processing unit 30a returns to the processing of step S303, changes the target captured image, and repeats the processing of steps S003 to S307.

  In step S309, the central processing unit 30a ends a series of automatic determination processes.

  As described above, the image processing apparatus 50 shown in FIG. 8 learns and recognizes the judgment tendency of the evaluator who is a specific individual, and performs the automatic discrimination process of the captured image based on the judgment reference value obtained by the learning process. It can be carried out. Therefore, the image processing apparatus 50 can perform automatic discrimination processing based on a judgment criterion that matches with high accuracy the judgment criteria required by the evaluator. The image processing device 50 has the same effect as the image determination device 10 shown in FIG.

DESCRIPTION OF SYMBOLS 10 ... Image judgment apparatus 12 ... Input device 14 ... Display apparatus 16 ... Memory | storage device 18 ... Image memory 20 ... Program memory 30, 30a ... Central processing unit 32 ... Judgment input process part 34 ... Image size acquisition part 36 ... Learning calculating part 38 Image quality adjustment unit 40 ... Feature extraction unit 42 ... Edge amount calculation unit 44 ... Data acquisition unit 46 ... Principal component analysis unit 50 ... Image processing device 52 ... Captured image determination unit 54 ... Comparison unit 60 ... Second feature extraction unit 62 ... Second edge amount calculation unit 64 ... second data acquisition unit

Claims (8)

A first storage unit for storing a learning image;
For the learning image, a feature extraction unit that acquires a specific value of an index related to image quality;
A display unit for displaying the learning image under a predetermined display condition;
An input unit for obtaining a judgment result of an evaluator for the learning image displayed by the display unit;
A second storage unit that stores a learning element that associates the determination result with the specific value;
From the learning element, a learning calculation unit that determines a judgment reference value of the evaluator regarding the index;
An image determination apparatus having The feature extraction unit includes a principal component analysis unit;
The principal component analysis unit obtains provisional concrete values corresponding to two or more provisional indices relating to image quality for each of the two or more learning images, performs principal component analysis on the provisional concrete values, and performs two or more provisional values. The image determination apparatus according to claim 1, wherein the specific value related to the principal component integrated with the provisional index is calculated.   The image determination apparatus according to claim 1, wherein the learning calculation unit performs a discriminant analysis on the learning element and determines the determination reference value.   The image viewing size selected by the evaluator is acquired, and the image viewing size is further reflected on the display condition, and further includes an image size obtaining unit. Image judging device. An image quality adjustment unit for adjusting the image quality of the learning image;
The image quality adjustment unit adjusts the image quality of the learning image so that the specific value of the index of the learning image becomes a predetermined first value;
The image determination apparatus according to claim 1, wherein the display unit displays the learning image adjusted by the image quality adjustment unit. An image determination device according to any one of claims 1 to 5,
A second specific value of a second index relating to image quality is acquired for the captured image, and the second specific value of the captured image is compared with the determination reference value determined by the image determination device. A photographed image judging unit for judging the photographed image;
An image processing apparatus.   A camera comprising the image processing apparatus according to claim 6. On the computer,
A procedure for reading the learning image stored in the first storage unit;
A procedure for calculating a specific value of an index related to image quality for the learning image;
Displaying the learning image under predetermined display conditions;
A procedure for obtaining a judgment result of an evaluator for the learning image displayed by the display unit;
Storing a learning element that associates the determination result with the specific value of the indicator;
An image determination program for executing, from the learning element, a procedure for determining a determination reference value of the evaluator regarding the index.

Images