JP6033724B2 - Image display system, server, and diagnostic image display device - Google Patents

Description

  The present invention relates to image quality evaluation technology, image generation technology, image display technology, and the like.

  There are two main methods for image quality evaluation. One is subjective evaluation, in which an image to be evaluated is presented to a large number of subjects to obtain an evaluation value. The other is objective evaluation, in which some feature amount is extracted from an image to be evaluated, and evaluation is performed based on the size.

  Subjective evaluation is an evaluation by human beings, and an evaluation highly correlated with human senses can be obtained. However, it is necessary to collect a large number of subjects, there are differences in the evaluation criteria of individual subjects, and in many evaluation target images. However, there are problems such as requiring a lot of time and cost.

  On the other hand, objective evaluation can be processed by a computer or the like, and the criteria are clear and a large number of evaluation results can be obtained in a short time. However, the results often deviate from human subjectivity.

  Therefore, a technology that aims to obtain objective evaluation results highly correlated with human subjectivity has been developed. For example, in Patent Document 1, as described in the summary, “a position on an image to which a subject pays attention when determining an overall image quality score is detected, and partial image information at the detected position is captured. Extracts image features and calculates image quality evaluation values according to the extracted features, and learns the relationship between the overall image quality score determined by the subject and the extracted image features and calculated image quality evaluation values. Based on the learning result, the position of the partial image to be evaluated for the image quality evaluation item necessary for determining the overall image quality score is specified for the evaluated image.The partial image at the specified position is captured. The image quality evaluation value for the image quality evaluation item necessary for determining the total image quality score for the captured partial image information is calculated. Calculating the total quality score for the evaluation image based on the results. "It has been disclosed.

Japanese Patent Laid-Open No. 10-063859

  However, since the above-described conventional technique uses one image, there are still problems such as variations in evaluation due to individual differences among subjects, and problems that time is required for gaze and evaluation time increases.

  The present invention has been devised in view of the above problems, and an object thereof is to provide an evaluation technique for more suitably obtaining an output result highly correlated with human subjectivity and an image generation processing technique using the evaluation technique.

  One embodiment of the present invention is an image display system including, for example, a server and a diagnostic image display device, and the server includes a communication unit capable of transmitting and receiving information to and from other devices via a network, and an image group. A diagnostic data display device, and an evaluation parameter determination unit that determines an evaluation parameter based on the image group, the diagnostic image display device, a communication unit capable of transmitting and receiving information to and from other devices via a network, An image input unit that inputs a diagnostic image, an image processing unit that performs image processing on the image input by the image input unit, and a display unit that displays an image subjected to image processing by the image processing unit, The image group stored in the database of the server is an image group previously ranked based on a predetermined evaluation index, and the evaluation parameter determination unit stores the image group in the database. The evaluation parameter is determined based on the ranking and feature amount of each image in the image group, the evaluation parameter determined from the communication unit of the server is transmitted to the communication unit of the diagnostic image display device, and the diagnostic image The image processing unit of the display device may be configured to perform the image processing on the image input by the image input unit so that the evaluation parameter received by the communication unit is more suitable.

  It is possible to provide an evaluation technique that more suitably obtains an output result having a high correlation with human subjectivity and an image generation processing technique using the evaluation technique.

It is a block diagram of Example 1 of the present invention. It is a block diagram of the evaluation parameter determination part in Example 1 of this invention. It is a flowchart of a feature-value production | generation part. It is a figure which shows the process of a feature-value production | generation part. It is a figure which shows the process in which the feature-value vector extraction part applies the feature-value. It is a figure which shows the process of vectorization of the feature-value vector extraction part. It is a flowchart in an evaluation parameter determination part. It is a block diagram at the time of connecting each part in an evaluation parameter determination part It is a block diagram of an image processing part It is a flowchart in an image processing part. It is a display example in a display part It is the 2nd example of a display in a display part. It is a block diagram of Example 3 of the present invention. It is a block diagram of the best shot image extraction part in Example 3 of this invention. It is a flowchart in the best shot image extraction part. It is a block diagram of Example 4 of the present invention. It is an example in the case of presenting two images in Example 4 of the present invention. It is an example in the case of showing more images than two sheets in Example 4 of the present invention.

  Embodiments will be described below with reference to the drawings.

  FIG. 1 is a block diagram of a learning server and a diagnostic image display device. In FIG. 1, a learning server 101 is connected to a diagnostic image display device 102 via a network 103. Here, the network 103 is a local area network (LAN) in a hospital, a metropolitan area network (MAN), a wide area network (WAN), the Internet, or the like. Further, the total number of diagnostic image display devices 102 connected to the network 103 may be an arbitrary number of 1 or more.

  The learning server 101 includes a database 104, an evaluation parameter determination unit 105, and a communication unit 106. In addition, there may be an input / output unit that performs input / output with respect to the database 104.

  The database 104 stores a group of images ranked in advance based on a predetermined evaluation index. Here, the order may include the same order. The predetermined evaluation index is, for example, “ease of viewing”, “beauty”, “ease of diagnosis”, “degree of disease progression”, and the like.

  For example, a questionnaire survey or the like is used to determine the ranking. At that time, two or more images for learning are presented to the subject, and their rank is collected as an answer.

  The evaluation parameter determination unit 105 determines an evaluation parameter from the stored image group.

  Details of the processing of the evaluation parameter determination unit 105 will be described with reference to FIG.

  FIG. 2 is a block diagram of the evaluation parameter determination unit 105. In FIG. 2, the feature amount generation unit 201 generates a feature amount from an image group stored in the database 104. The feature quantity vector extraction unit 202 uses the feature quantity generated by the feature quantity generation unit 201 to extract a feature quantity vector. The evaluation parameter calculation unit 203 calculates an evaluation parameter that appropriately associates the magnitude relationship between the pre-assigned rank and the feature quantity vector.

  First, the feature value generation unit 201 will be described. The feature amount generation unit 201 generates a feature amount from the image group using unsupervised learning. FIG. 3 shows a flowchart of the process of the feature quantity generation unit 201. The flow of processing is shown in FIG.

  In step 301, rectangular patches of a certain size, for example, 6 × 6 pixels, are randomly extracted from all the images in the database 104. Note that the number and shape of patches to be extracted may be arbitrary. For example, tens of thousands of patch groups 401a are extracted. Next, the brightness and contrast of each patch are normalized. Note that examples of normalization processing include, for example, a method of making the minimum and maximum values in a patch correspond to 0 to 1, a method of converting to a logarithm, and the like.

  Next, in step 302, an arbitrary number, for example, 100 representative patch groups 402a are calculated from the patch group 401a using unsupervised learning. Examples of the unsupervised learning method used at this time include a k-means method, a neural network, and deep learning.

  The processes in steps 301 and 302 described above are performed an arbitrary number of times, for example, 10 times. Representative patch groups 402a to 402j are obtained by 10 calculations.

  In step 304, unsupervised learning is again applied to the 1000 representative patch groups 402a to 402j to obtain an arbitrary number, for example, 200 representative patch groups 403. In the present invention, the representative patch group 403 is referred to as a feature amount.

  The reason for performing unsupervised learning in two stages is that when unsupervised learning is applied to a huge number of patches (for example, 10 million), the solution becomes unstable.

  Next, the feature vector extraction unit 202 will be described. The feature amount vector extraction unit 202 extracts a feature amount vector for each image stored in the database 104 from the feature amount generated by the feature amount generation unit 201.

  For a certain feature quantity extraction target image 501, as shown in FIG. 5, for the k-th patch of the feature quantity 403, the calculation shown in the following Equation 1 is performed.

Here, the vector x is a rectangular area in the feature quantity extraction target image 501 having the same size as the representative patch, and the vector c ^(k) is the ^{kth (} 200th representative patch generated in the first step ⁾ . k = 1,2, ..., 200). μ (z) is the average of z _k with respect to k. This process is performed at all pixel positions of the feature quantity extraction target image 501.

When Expression (1) is performed for all k, 200 processing results f _k of the same size are obtained for one feature quantity extraction target image 501.

Here, a pooling process for generating a feature vector from these 200 processing results will be described. In the following, a method called Sum-pooling will be described. For example, as shown in FIG. 6, each is divided into four equal parts in the vertical and horizontal directions, and the sum is obtained in each of the areas 601 to 604. The sum of the regions 601 is set as the first dimension of the feature vector, the sum of the regions 602 is set as the value of the second dimension, and the region 603 is set as the third dimension. In this manner, all 200 f _k are converted into a one-dimensional vector, which is used as a feature vector. Thereby, a 4 × 200 = 800-dimensional feature vector is obtained for one image. Note that the number of area divisions does not necessarily have to be four. The minimum is 1 (no division), and the maximum division size matches the image size. In the above-described processing, the sum calculation is performed after the division. However, a feature amount vector may be generated using a method (max-pooling) for obtaining a maximum value in each divided region.

  In addition, for the feature value generation and feature value vector extraction unit 202 in the feature value generation unit 201, one of existing feature values or a combination thereof is used in addition to the above-described using unsupervised learning. May be. Examples of feature quantities to be used include histograms, edge feature quantities represented by HOG (Histograms of Oriented Gradients), local feature quantities such as SIFT (Scale-Invariant Feature Transform), and feature quantities by sparse coding.

  Next, the evaluation parameter calculation unit 203 will be described. The evaluation parameter calculation unit 203 calculates a parameter vector using Ranking SVM (Support Vector Machine) for the ranked image group.

  For two images i and j, i is higher in rank than j for a certain evaluation index. The feature quantity vectors extracted from the respective images by the feature quantity vector extraction unit 202 are assumed to be xi and xj.

  Here, calculating the parameter vector w satisfying the relationship shown in the following Equations 2 and 3 is called learning in the present invention.

SMO (sequential minimal optimization) is often used for SVM optimization. When the number of samples is n, SMO is an efficient optimization method that can be solved with O (n ² ), but Ranking SVM used in the present invention learns about the order of the set of two images. Therefore, the maximum number of samples is _n C ₂ , which is easily affected by the number of images.

  Therefore, in the present invention, an online learning optimization method is introduced, and a new optimization method that enables optimization with the amount of computation of O (n) is used.

  FIG. 7 is a flowchart showing the processing of the evaluation parameter calculation unit 203. In step 702 and 703, the presence / absence of the order of the two images in the predetermined evaluation index is determined for the feature vector extracted from the set of two images. If there is an order relationship, an error d in the case where there is an order relationship is calculated in step 705. If there is no order relation, an error d in the case where there is no order relation is calculated in step 706. In step 707, an error d at the time of update is determined. If the error d is large, w is updated in step 708, and the same processing is performed by changing the set of images. When the error d becomes sufficiently small, the learning is finished.

Hereinafter, the update formula will be described in detail.

  A set of sets having an order relationship in an image group is O, and a set of sets having the same order relationship is S. Here, the concept of soft margin maximization is introduced, and an objective function to be optimized is defined as in the following Equation 4.

In this equation, α and β are slack variables of the set O and the set S, respectively, and are 0 or more in each set. C is a positive hyperparameter that balances the regularization term.

  Here, since the set {i, j} of two images taken out from the set O having the order relationship is not included in the set having the same order relationship, the first term of Equation 4 is expressed as O And S can be solved independently.

  Therefore, first, consider solving the image set {i, j} included in the set O having the order relation calculated in step 705. First, a hinge loss function is given for {i, j} as shown in Equation 5.

Now, instead of Equation 4, optimization is performed sequentially using Equation 6 below.

In this equation, ξ is a slack variable, and ξ ≧ 0. C is a positive hyperparameter that balances the regularization term. The parameter vector w is obtained by solving Equation 6 using constrained quadratic programming.
Equation 6 is solved under the Karush-Kuhn-Tucker (KKT) condition, for example. First, a result like Formula 7 is obtained.

When w is obtained, Equation 9 is obtained through Equation 8.

  Note that λ is given by Equation 10 below.

  Therefore, the error d is expressed by Equation 11.

  The update formula can be solved as shown in Formula 12.

  On the other hand, consider the case of a set of images {i, j} taken out from a set S of sets having the same order relation calculated in step 706.

  When a hinge loss function is given as in the case where there is an order relationship, Equation 13 is obtained.

Here, ε is an allowable error. The objective function used for optimization can be expressed as in the following Expression 14, similarly to Expression 6.

The parameter vector w is obtained by solving Equation 14 using a constrained quadratic programming method.

  The above equation is solved under the KKT condition, for example, as in the case of the set O having the order relation. First, Formula 15 is obtained.

Here, sgn is a sign function.

  Next, when w is calculated, Equation 16 is obtained.

Note that λ is given by Equation 17 below.

  Therefore, the error d is expressed by Equation 18.

  Therefore, the update equation can be solved as in Equation 19.

  As described above, the final parameter vector w calculated by sequentially updating is used as an evaluation parameter for a certain evaluation index. The communication unit 106 transmits the calculated evaluation parameter to the diagnostic image display device 102. In addition, the communication unit 106 transmits the feature quantity generated by the feature quantity generation unit 201 to the diagnostic image display apparatus 102 so that the diagnostic image display apparatus 102 can also calculate the feature quantity vector of the input image.

  Further, in the learning server 101, as shown in FIG. 8, the feature quantity generation unit 201, the feature quantity vector extraction unit 202, and the evaluation parameter calculation unit 203 in the evaluation parameter determination unit 105 are connected to the database 104, and the generated feature quantity and The extracted feature vector may be stored in the database 104 and used in the second and subsequent calculations.

  Note that the learning server 101 may be realized as a program executed by a computer. The computer has an input / output device, a CPU (Central Processing Unit), a memory, a recording medium, and the like, and these are connected to a bus. The database corresponds to a recording medium of the computer. The recording medium stores a program for realizing the server. The program is expanded in the memory through the bus and executed by the CPU.

  Next, the operation of the diagnostic image display apparatus 102 will be described.

  As shown in FIG. 1, the diagnostic image display apparatus 102 includes a communication unit 107, an image input unit 108, an image processing unit 109, and a display unit 110.

  The diagnostic image display apparatus 102 generates a plurality of processing results for the image input from the image input unit 108, and displays a more preferable processing result using the feature amount and the evaluation parameter received from the learning server 101.

  The configuration of the image processing unit 109 is shown in FIG. The image processing unit 109 includes a processed image generation unit 901, a feature vector extraction unit 902, an evaluation value calculation unit 903, and a selection unit 904. Each process will be described with reference to FIG.

  In FIG. 10, as step 1001, an image is input from the image input unit 108. Next, in step 1002, the processed image generation unit 901 generates a plurality of color correction processing results based on a plurality of color patterns, for example.

  Next, in step 1003, the feature quantity vector extraction unit 902 uses the feature quantities received from the learning server 101 to extract feature quantity vectors for each of a plurality of processing results. The extraction method is the same as that of the feature vector extraction unit 202 of the learning server 101.

  Next, in step 1004, the evaluation value calculation unit 903 calculates a weighted linear sum of the extracted feature vector and the evaluation parameter received from the learning server 101, and calculates an evaluation value for each processing result.

  Finally, in step 1005, the determination unit 904 ranks the processing results using the evaluation values, and selects the processing result having the maximum evaluation value.

  The display unit 110 may simply use the image having the maximum evaluation value selected by the determination unit 904 as the final output image. Further, as described below, an image having the maximum evaluation value selected by the determination unit 904 may be displayed, and a plurality of color correction processing result images serving as other final output image candidates may be displayed to the user. Good.

  FIG. 11 is an example of a screen displayed to the user on the display unit 110. The processing result 1101 selected by the determination unit 904 and having the maximum evaluation value is displayed. Further, the other images 1102a to 1102d for which the evaluation values are calculated are displayed side by side. The number of displayed sheets may be arbitrary, and it is desirable that the number can be selected by the user. Moreover, as a display location, the lower part of a screen as shown in FIG. 11, the screen side as shown in FIG. 12, etc. can be considered.

  In addition, a keyboard, mouse, button, touch panel, or the like may be used so that 1102a to d can be displayed in a large size. If there are images other than 1102a to 1102d, scrolling or the like may be performed, and an image that cannot be displayed may be browsed.

  As described above, according to the present embodiment, it is possible to output a correction result having a high correlation with the subjectivity using the relative evaluation of the rank considered to have a small individual difference.

  Next, a second embodiment of the present invention will be described. In this embodiment, it is possible to learn a plurality of evaluation indexes at a time. In the present embodiment, the process of the evaluation parameter calculation unit 203 in FIG. 2 is changed as follows. Other configurations and operations are the same as those in the first embodiment, and thus description thereof is omitted.

  For the two images i and j ranked for a plurality of evaluation indexes, the feature quantity vectors extracted from the respective images by the feature quantity vector extraction unit 202 are assumed to be xi and xj.

  In the present embodiment, two indices are randomly extracted from a plurality of evaluation indices that are ranked in advance, and a parameter vector is obtained. By repeating this optimization process, a plurality of evaluation indices can be learned at once.

Here, it is considered that the two evaluation indexes w ^u and w ^v are optimized in the two images i and j.

The two evaluation indexes w ^u and w ^v may be evaluation indexes having an order relationship with respect to i and j, or may be evaluation indexes having the same order relationship. Accordingly, regarding possible combinations, that is, the order relationship, two cases have an order relationship, one case has an order relationship and the other has the same order relationship, and the two have the same order relationship.

  First, the case where the two have an order relationship will be described.

  Here, the hinge loss function is given as in Equation 20 below.

Next, parameter vectors w ^u and w ^v are obtained by sequentially optimizing with the following equation (21). As in the first embodiment, a soft margin is introduced.

The parameter vector can be calculated by optimizing Equation 21.

Next, a case where one side has an order relationship and the other has the same order relationship will be described. Here, it is assumed that the order relation exists for w ^{u and the} order relation for w ^v is equal. If the order relationship is equal for w ^{u and the} order relationship exists for w ^v , the evaluation indices w ^u and w ^v may be switched.

  Here, the hinge loss function is given by the following formula 22.

The parameter vectors w ^u and w ^v are obtained by sequentially optimizing with the following equations 23 and 24. As in the first embodiment, a soft margin is introduced.

In Equations 23 and 24, ν is a slack variable, and ν ≧ 0. D is a positive hyperparameter that balances the regularization term. The parameter vectors w ^u and w ^v can be calculated by optimizing the equations (23) and (24).

  Next, the case where the two are in the same order relationship will be described. Similarly, the hinge loss function is given by the following formula 25.

The parameter vectors w ^u and w ^v are obtained by sequentially optimizing with the following equation (26). As in the first embodiment, a soft margin is introduced.

In this equation, ξ is a slack variable, and ξ ≧ 0. C is a positive hyperparameter that balances the regularization term. The parameter vectors w ^u and w ^v can be calculated by optimizing the above equation.

  As described above, according to the present embodiment, it is possible to learn a plurality of evaluation indexes at once.

  Next, as Example 3 of the present invention, an example will be described in which when a video (moving image) is input, a best shot image is extracted from a plurality of frames having different imaging times in the video. Each frame of the video is evaluated at each time, and a frame exceeding the threshold is displayed as a best shot image. FIG. 13 is a block diagram in the present embodiment. The image processing unit 109 in FIG. 1 is changed to a best shot image extraction unit 1301. The other components having the same reference numerals as those in the first embodiment are the same as those in the first embodiment, and thus the description thereof is omitted. The configuration of the best shot image extraction unit 1301 is shown in FIG. The best shot image extraction unit 1301 includes a frame extraction unit 1401, a feature vector extraction unit 902, an evaluation value calculation unit 903, and a threshold processing unit 1402. The processing of the best shot image extraction unit 1301 will be described with reference to FIG.

  First, in step 1501, an image is input. Next, in step 1502, the frame extraction unit 1401 extracts a frame from the input video. Next, in step 1503, a feature amount vector extraction unit 902 extracts a feature amount vector from the extracted frame. Next, in step 1504, the evaluation value calculation unit 903 calculates an evaluation value. Finally, in step 1505, the threshold processing unit 1402 compares the calculated evaluation value with a preset threshold, and if the threshold is exceeded, the frame is selected. The selected image is displayed on the display unit 110.

  Also, instead of the process of selecting and displaying the first frame whose evaluation value exceeds the threshold value, the frame selection process is repeated for frames within the target time range, and a plurality of selection results are held, May be finally displayed from among the selected frames after the input to the end of the target time range.

  As described above, according to the present embodiment, it is possible to extract the best shot based on the evaluation index from the video.

  Next, Embodiment 4 of the present invention will be described.

  In the diagnostic image display apparatus 102, a form in which a plurality of generated processing results are presented to the user and answer results are collected will be described. In this embodiment, the evaluation index for each user can be learned individually, and an output having a higher correlation with the user's evaluation index can be obtained.

  Example 4 will be described with reference to the drawings. FIG. 16 is a block diagram showing a configuration in the present embodiment. In the diagnostic image display apparatus 102, the image processing unit 109 is changed to a processed image generation unit 901. In addition, an answer input unit 1601 for acquiring the user's answer is added. The answer input unit 1601 is connected to the communication unit 107. In addition, about the part which attached | subjected the same code | symbol, it is the same as the above-mentioned Example, and abbreviate | omits description.

  The display unit 110 presents a plurality of processing results generated by the processed image generation unit 901, and performs a display that prompts the user to determine the rank of a predetermined evaluation index. For example, a display example in the case of presenting the evaluation index “cuteness” is shown in the figure. FIG. 17 shows a presentation example in the case of two sheets. FIG. 18 shows an example of presentation when three images are presented. In any case, together with a plurality of images, a display that prompts the user to determine the rank of the evaluation index “cuteness” is performed.

  The answer input unit 1601 acquires the selection result of the user according to the above display. As an acquisition method, selection with a keyboard, mouse, button, touch panel, or the like can be considered. The communication unit 107 transmits the answer result and the presented image to the learning server 101.

  The learning server 101 records the answer result and the presentation image received by the communication unit 106 in the database 104. The recorded answer results and presentation images are used when the evaluation parameter determination unit 105 next performs learning.

  The timing for performing the relearning may be when a new answer result is obtained or may be determined at an interval. In order to obtain an optimal result in real time, it is desirable to perform learning whenever a new answer result is obtained. However, since learning generally requires a great deal of time, it is possible to reduce the number of re-learning operations and to shorten the calculation time by collectively learning the response results recorded during a certain period.

  As described above, according to the present embodiment, the diagnostic image display apparatus collects the user's answer and reflects it in the learning result, thereby obtaining an output having a higher correlation with the user's evaluation index.

  In the foregoing, a plurality of embodiments of the present invention have been described. A plurality of these embodiments may be carried out independently or in combination. In other words, the learning process disclosed in the second embodiment may be installed in the learning servers of the third and fourth embodiments. Further, the best shot image extraction unit of the third embodiment may be mounted on the diagnostic image display devices of the first, second, and fourth embodiments. In addition, the configuration of the display unit and the answer input unit that perform display for prompting the user to determine the order of the predetermined evaluation index in the fourth embodiment may be applied to the first, second, and third embodiments.

101 Learning Server 102 Diagnostic Image Display Device 103 Network 104 Database 105 Evaluation Parameter Determination Unit 106 Communication Unit (Server Side)
107 Communication unit (diagnostic image display device side)
108 image input unit 109 image processing unit 110 display unit 201 feature quantity generation unit 202 feature quantity vector extraction unit 203 evaluation parameter calculation unit 401 extracted patch group 402 representative patch group 403 by first unsupervised learning feature quantity 501 feature quantity Extraction target image 901 Processed image generation unit 902 Feature quantity vector extraction unit 903 Evaluation value calculation unit 904 Determination unit 1301 Best shot image extraction unit 1401 Frame extraction unit 1402 Threshold processing unit 1601 Answer input unit

Claims (13)

An image display system including a server and a diagnostic image display device,
The server
A communication unit capable of transmitting and receiving information to and from other devices via a network;
A database for storing images;
An evaluation parameter determination unit for determining an evaluation parameter based on the image group;
With
The diagnostic image display device includes:
A communication unit capable of transmitting and receiving information to and from other devices via a network;
An image input unit for inputting a diagnostic image;
An image processing unit that performs image processing on the image input by the image input unit;
A display unit for displaying an image subjected to image processing by the image processing unit,
The image group stored in the database of the server is an image group ranked in advance based on a predetermined evaluation index,
The evaluation parameter determination unit
Generating a feature quantity by unsupervised learning for the image group stored in the database, and extracting a feature quantity vector from each image of the image group using the generated feature quantity;
For the set of images including a plurality of images in the image group, depending on whether or not there is an order relationship of the set of images, by updating the evaluation parameter using different calculation formulas, Calculating the evaluation parameter that associates a feature vector with a rank based on a predetermined evaluation index in the set of images;
Transmitting the evaluation parameter calculated from the communication unit of the server to the communication unit of the diagnostic image display device;
The image processing unit of the diagnostic image display device performs the image processing on the image input by the image input unit so that the evaluation parameter received by the communication unit is more suitable. Display system.   The image display system according to claim 1,
  When the ranking based on a plurality of different evaluation indexes is made in the image group stored in the database, the evaluation parameter determination unit of the server uses two of the plurality of different evaluation indexes for each set of images. An image display system, wherein an index is extracted, and the evaluation parameter for each of the plurality of different evaluation indices is calculated at a time by updating the evaluation parameter based on the two extracted indices.   The image display system according to claim 2,
  The image group stored in the database of the server is an image transmitted from the communication unit of the diagnostic image display device and received by the communication unit of the server together with information on the order generated in the diagnostic image display device. An image display system comprising:   The image display system according to claim 3,
  The information about the ranks generated in the diagnostic image display device is such that the display unit of the diagnostic image display device prompts the user to display a plurality of images and to determine the rank of a predetermined evaluation index for the plurality of images. An image display system that performs display and is generated based on an input result input by a user answer input unit included in the diagnostic image display device.   A server connectable with a diagnostic image display device via a network,
    A communication unit capable of transmitting and receiving information to and from the diagnostic image display device via a network;
    A database for storing images;
    An evaluation parameter determination unit for determining an evaluation parameter based on the image group;
  With
  The image group stored in the database is an image group that is ranked in advance based on a predetermined evaluation index,
  The evaluation parameter determination unit
  Generating a feature quantity by unsupervised learning for the image group stored in the database, and extracting a feature quantity vector from each image of the image group using the generated feature quantity;
  For the set of images including a plurality of images in the image group, depending on whether or not there is an order relationship of the set of images, by updating the evaluation parameter using different calculation formulas, Calculating the evaluation parameter that associates a feature vector with a rank based on a predetermined evaluation index in the set of images;
  The server, wherein the evaluation parameter calculated from the communication unit of the server is transmitted to the communication unit of the diagnostic image display device.   The server according to claim 5, wherein
  The evaluation parameter determination unit extracts two indexes from the plurality of different evaluation indexes for each set of images when ranking based on a plurality of different evaluation indexes is performed in the image group stored in the database. And updating the evaluation parameters based on the two extracted indices to calculate each evaluation parameter for the plurality of different evaluation indices at a time.   The server according to claim 6, wherein
  The server, wherein the image group stored in the database includes images transmitted from the diagnostic image display device and received by the communication unit, together with information related to the order generated in the diagnostic image display device.   A diagnostic image display device connectable with a server via a network,
  A communication unit capable of transmitting and receiving information to and from the server via a network;
  An image input unit for inputting a diagnostic image;
  An image processing unit that performs image processing on the image input by the image input unit;
  A display unit for displaying an image subjected to image processing by the image processing unit;
  With
  The image processing unit performs a plurality of color correction processes on the image input by the image input unit to generate a plurality of processed images, and based on the feature amount received from the server via the communication unit Calculating each feature amount vector of the plurality of processed images, calculating an evaluation value of each of the plurality of processed images in an evaluation parameter received from the server based on the feature amount vector, and The processed image with the best evaluation value is output as an image to be displayed on the display unit.
  A diagnostic image display device characterized by that.   The diagnostic image display device according to claim 8,
  The communication unit is
  The display unit displays a plurality of images and displays prompting the user to determine the order of predetermined evaluation indexes for the plurality of images, and an input result input by a user answer input unit provided in the diagnostic image display device A diagnostic image display device that transmits information related to a rank in a predetermined evaluation index for the plurality of images generated based on the image to the server.   An image display system including a server and a diagnostic image display device,
  The server
    A communication unit capable of transmitting and receiving information to and from other devices via a network;
    A database for storing images;
    An evaluation parameter determination unit that determines an evaluation parameter based on the image group;
  With
  The diagnostic image display device includes:
    A communication unit capable of transmitting and receiving information to and from other devices via a network;
    An image input unit to which a moving image including a plurality of frames having different imaging times is input;
    An image extraction unit for extracting a predetermined image from the plurality of frames for the moving image input by the image input unit;
    A display unit for displaying the image extracted by the image extraction unit;
  With
  The image group stored in the database of the server is an image group ranked in advance based on a predetermined evaluation index,
  The evaluation parameter determination unit
  Generating a feature quantity by unsupervised learning for the image group stored in the database, and extracting a feature quantity vector from each image of the image group using the generated feature quantity;
  For the set of images including a plurality of images in the image group, depending on whether or not there is an order relationship of the set of images, by updating the evaluation parameter using different calculation formulas, Calculating the evaluation parameter that associates a feature vector with a rank based on a predetermined evaluation index in the set of images;
  Transmitting the evaluation parameter calculated from the communication unit of the server to the communication unit of the diagnostic image display device;
  The image extraction unit of the diagnostic image display apparatus extracts the plurality of frames of the moving image input by the image input unit by calculating and comparing evaluation values in the evaluation parameters received by the communication unit, respectively. Determine the image to be used
  An image display system characterized by that.   The image display system according to claim 10,
  When the ranking based on a plurality of different evaluation indexes is made in the image group stored in the database, the evaluation parameter determination unit of the server uses two of the plurality of different evaluation indexes for each set of recording images. An image display system, wherein an evaluation parameter is extracted for each of the plurality of different evaluation indices at a time by extracting an index and updating the evaluation parameter based on the two extracted indices.   The image display system according to claim 11,
  The image group stored in the database of the server is an image transmitted from the communication unit of the diagnostic image display device and received by the communication unit of the server together with information on the order generated in the diagnostic image display device. An image display system comprising:   The image display system according to claim 12,
  The information about the ranks generated in the diagnostic image display device is such that the display unit of the diagnostic image display device prompts the user to display a plurality of images and to determine the rank of a predetermined evaluation index for the plurality of images. An image display system that performs display and is generated based on an input result input by a user answer input unit included in the diagnostic image display device.

Images