JP2018161206A - Image processing apparatus, image processing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve identification accuracy.SOLUTION: An image processing apparatus 100 includes an image acquisition part 11, an image database 21, a similarity rank calculation part 13, an estimation part 14, a determination part, and a presentation part 16. The image acquisition part 11 acquires image data. The image database 21 stores the image data in association with a name relating to the image data. The similarity rank calculation part 13 calculates a first rank of the similarity of the respective image data in the image database 21, with the image data acquired by the image acquisition part 11. The estimation part 14 estimates a name relating to the image data acquired by the image acquisition part 11. The determination part determines whether or not the name associated with the first-rank image data calculated by the similarity rank calculation part 13, accords with the name estimated by the estimation part 14. The presentation part 16 performs a prescribed presentation according to a result determined by the determination part.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an image processing apparatus, an image processing method, and a program.

  Techniques have been developed for assisting diagnosis by imaging an affected area to be diagnosed and processing the captured image. For example, Patent Document 1 describes a similar image search device that can perform a similar image search satisfactorily even between images having different image formats and the like.

JP 2013-200590 A

  The similar image search device described in Patent Literature 1 can perform a similar image search better than the conventional technique. However, there are a certain number of cases where the disease names are different even if the images are similar, and if the conventional technology is applied to such cases, an image with an incorrect disease name is searched as a similar image. . As a result, there is a problem that the identification accuracy of the disease name is lowered.

  SUMMARY An advantage of some aspects of the invention is that it provides an image processing apparatus, an image processing method, and a program capable of improving identification accuracy.

In order to achieve the above object, an image processing apparatus of the present invention provides:
An image acquisition unit for acquiring image data;
An image database for storing the image data and a name related to the image data in association with each other;
The similarity indicating the degree of similarity to the image data acquired by the image acquisition unit is obtained for each image data in the image database, and the first rank of the similarity of the image data in the image database is calculated. A similarity ranking calculation unit,
An estimation unit that estimates a name related to the image data acquired by the image acquisition unit;
A determination unit that determines whether a name associated with the first-order image data calculated by the similarity rank calculation unit matches a name estimated by the estimation unit;
A presentation unit that performs a predetermined presentation according to a result determined by the determination unit;
Is provided.

  According to the present invention, identification accuracy can be improved.

It is a figure which shows the function structure of the image processing apparatus which concerns on embodiment of this invention. It is a figure which shows an example of the data memorize | stored in the image database of the image processing apparatus which concerns on embodiment. It is a figure which shows an example of the data identification by linear SVM. It is a flowchart of the image database registration process of the image processing apparatus which concerns on embodiment. It is a flowchart of the estimation part learning process of the image processing apparatus which concerns on embodiment. 3 is a flowchart of image processing of the image processing apparatus according to the embodiment. It is a figure explaining the mode of the correction | amendment of the order by the order correction | amendment part of the image processing apparatus which concerns on embodiment. It is a figure explaining the mechanism of the correction | amendment of the order by the order correction | amendment part of the image processing apparatus which concerns on embodiment.

  Hereinafter, an image processing apparatus, an image processing method, and a program according to embodiments of the present invention will be described with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals.

  When image data whose disease name is unknown is input, the image processing apparatus according to the embodiment of the present invention searches the image database for image data similar to the image data, and ranks them in order of high similarity. Then, using the classification result obtained by the estimation unit learned based on the image data and the disease name, the similarity ranking is corrected as necessary, and stored in the image database based on the corrected ranking. Present image data together with the disease name. Thereby, the user can estimate the disease name of the image data input to the image processing apparatus. The functional configuration and processing contents of such an image processing apparatus will be described below.

  As illustrated in FIG. 1, the image processing apparatus 100 according to the embodiment includes a control unit 10, a storage unit 20, an image input unit 31, a name input unit 32, and an output unit 33.

  The control unit 10 is configured by a CPU (Central Processing Unit) or the like, and by executing a program stored in the storage unit 20, each unit (image acquisition unit 11, feature vector extraction unit 12, similarity rank calculation unit 13) described later. The functions of the estimation unit 14, the rank correction unit 15, and the presentation unit 16) are realized.

  The storage unit 20 includes a ROM (Read Only Memory), a RAM (Random Access Memory), and the like, and stores a program executed by the CPU of the control unit 10 and necessary data. The storage unit 20 functionally includes an image database 21.

  As shown in FIG. 2, the image database 21 includes the image data input from the image input unit 31, the feature vector of the image data extracted by the feature vector extraction unit 12, and the image data input from the name input unit 32. It is a database that stores disease names linked to each other.

  The image input unit 31 includes a camera equipped with a special magnifying glass called a dermoscope, captures an image of an affected area of a disease, and inputs the captured image data.

  The name input unit 32 includes a keyboard, a mouse, a touch panel, and the like. When image data of an affected part whose disease name (disease name) is known is input from the image input unit 31, the name of the disease is input. Accept.

  The output unit 33 includes a display such as an LCD (Liquid Crystal Display), an organic EL (Electro-Luminescence) display, and the like, an image input from the image input unit 31, a disease name estimated from the image, and similar to the image The images in the image database 21 are displayed.

  Next, a functional configuration of the control unit 10 of the image processing apparatus 100 will be described. The control unit 10 implements the functions of an image acquisition unit 11, a feature vector unit 12, a similarity rank calculation unit 13, an estimation unit 14, a rank correction unit 15, and a presentation unit 16.

  The image acquisition unit 11 acquires image data of a diseased part from the image input unit 31. When the name input unit 32 receives an input of the disease name of the diseased part, the image acquisition unit 11 also acquires the disease name. Then, when registering the acquired image data and disease name in the image database 21, the image acquisition unit 11 associates the image data with the disease name and registers them in the image database 21.

  The feature vector extraction unit 12 extracts feature vectors from the image data acquired by the image acquisition unit 11. The type of the feature vector is arbitrary, but, for example, a feature vector in a BoVW (Bag of Visual Words) format is extracted. Then, when registering the extracted feature vector in the image database 21, the feature vector extracting unit 12 registers the feature vector in association with the image data of the extraction source.

  Using the feature vector extracted by the feature vector extraction unit 12, the similarity rank calculation unit 13 stores each similarity stored in the image database 21 with a degree of similarity indicating the degree of similarity to the image data to be searched (query image data). Find image data. Then, the similarity rank calculation unit 13 calculates the rank in the image data stored in the image database 21 of the obtained similarity of each image data. Specifically, first, the feature vector extraction unit 12 extracts a feature vector of search target image data (query image data). Then, the similarity rank calculation unit 13 calculates the similarity between the feature vector and each feature vector stored in the image database 21, and sorts the image data stored in the image database 21 in descending order of the similarity. To do. Then, by assigning ranks to the sorted results in order, the similarity rank calculation unit 13 determines the similarity of each image data stored in the image database 21 with the image data to be searched (query image data). The rank of height can be obtained.

  The estimation unit 14 estimates the disease name of the diseased part based on the feature vector of the image data of the diseased part. The estimation unit 14 performs this estimation with a discriminator using, for example, a linear SVM (Support Vector Machine). The linear SVM is to determine the boundary between two types (two classes) of data and identify the type of data (class). For example, as shown in FIG. 3, when S data and ☆ data are given, the linear SVM determines its boundary line L. Thereby, the linear SVM can identify whether the data belongs to the class of ○ or the data belonging to the class of ☆ even for unknown data.

  One linear SVM can only identify two classes, but by preparing n linear SVMs (i = 1 to n) for identifying the i-th class and the non-i-th class, Identification can be made. In this case, for unknown data, the i-th class that the linear SVM with the highest score (the distance between the data and the boundary line is the maximum) identifies from other classes is the identification result of the data. It becomes. Such a configuration composed of a plurality of linear SVMs so that many classes can be classified is called a linear multi-class SVM.

  Note that the estimation unit 14 can be realized by using not only a linear SVM classifier but also a nonlinear SVM classifier, a neural network, or the like. A linear SVM separates two classes at a linear boundary (straight line, plane, hyperplane), whereas a non-linear SVM separates two classes at a non-linear boundary. However, since the nonlinear SVM is such that data in real space is once mapped to a space that can be separated at a linear boundary and then separated by the same mechanism as the linear SVM, the essential part may be considered the same as the linear SVM. Therefore, the linear SVM and the nonlinear SVM are simply referred to as SVM. In addition, the non-linear SVM can constitute a non-linear multi-class SVM similarly to the linear SVM. The linear multi-class SVM and the non-linear multi-class SVM are collectively referred to as a multi-class SVM.

  Further, in the SVM, n or more SVMs are required for identification of n (n> 2) classes or more. However, when a neural network is used, three or more classes can be identified at a time. However, the neural network has a drawback in that the discrimination performance cannot be improved if the number of learning samples is not greater than a certain number, as compared with the SVM. In any case, when the estimation unit 14 inputs a feature vector of unknown image data by performing identification using linear SVM, nonlinear SVM, neural network, or the like based on the learning data used at the time of learning, the above-described mechanism is used. The most likely disease name can be estimated from the feature vector.

  The rank correction unit 15 calculates the rank of similarity between the image data calculated by the similarity rank calculation unit 13 and the image data (query image data) to be searched for, stored in the image database 21. Correction based on the estimation result by Specifically, the disease name associated with the image data having the first similarity with the image data to be searched (query image data) does not match the estimation result by the query image data estimation unit 14. In this case, assuming that the estimation result of the estimation unit 14 is correct, the similarity ranking of the image data associated with the disease name that matches the estimation result is increased by N (for example, 1).

  The presentation unit 16 outputs the image data and the disease name from the image data having the first similarity to the predetermined order (for example, up to the fifth), based on the order corrected by the order correction unit 15. To display. In addition, the presentation method of the result by the presentation part 16 can be changed suitably. For example, only the image data with the first similarity may be displayed on the output unit 33, or only the disease name associated with the image data with the first similarity may be displayed on the output unit 33. . Further, the presentation unit 16 calculates, for example, the disease name calculated by the similarity rank calculation unit 13 and associated with the image data having the first similarity, the estimation result by the query image data estimation unit 14, May be displayed on the output unit 33. That is, the presentation unit 16 matches the disease name calculated by the similarity ranking calculation unit 13 and associated with the image data having the first similarity with the estimation result of the query image data estimation unit 14. A predetermined presentation is made according to the determination result of whether or not to do.

  The functional configuration of the image processing apparatus 100 has been described above. Next, a process in which the image processing apparatus 100 registers image data in the image database 21 will be described with reference to FIG. This process is started when the user instructs the image processing apparatus 100 to register image data in the image database 21.

  First, the image acquisition unit 11 of the image processing apparatus 100 acquires image data from the image input unit 31 (step S101). Next, the image acquisition unit 11 acquires a disease name from the name input unit 32 (step S102). And the feature vector extraction part 12 extracts a feature vector from the image data which the image acquisition part 11 acquired by step S101 (step S103).

  Then, the control unit 10 associates the image data acquired by the image acquisition unit 11 in step S101, the disease name acquired by the image acquisition unit 11 in step S102, and the feature vector extracted by the feature vector extraction unit 12 in step S103. Then, it is registered in the image database 21 (step S104). And the control part 10 determines whether registration to the image database 21 was complete | finished about all the images input from the image input part 31 (step S105). If not completed (step S105; No), the process returns to step S101. If completed (step S105; Yes), the image database registration process is terminated.

  Through the above image database registration processing, image data, feature vectors, and disease names are registered in association (linked) with the image database 21 as shown in FIG. Next, an estimation unit learning process for allowing the estimation unit 14 of the image processing apparatus 100 to learn and estimating a disease name from unknown image data will be described with reference to FIG. This process is started when the user instructs the image processing apparatus 100 to learn the estimation unit 14. Here, the image data and the disease name registered in the image database 21 are used as learning data for causing the estimation unit 14 to learn. However, the image data and the disease name for learning data are separately prepared, and the following estimation is performed. It may be used in the part learning process.

  First, the control unit 10 of the image processing apparatus 100 selects one image data from the image database 21. For this selection, for example, the image data stored in the image database 21 is selected one by one, such as selecting in the order registered in the image database 21, and finally all the image data can be selected. If so, they may be selected in any order. And the control part 10 acquires the feature vector and disease name linked | related with the selected image data as learning data (step S201). Next, the control unit 10 causes the estimation unit 14 to learn using the feature vector and disease name acquired in step S201 (step S202). And the control part 10 determines whether the learning to the estimation part 14 was complete | finished about all the image data memorize | stored in the image database 21 (step S203). If not completed (step S203; No), the process returns to step S201, and if completed (step S203; Yes), the estimation unit learning process is terminated.

  Through the above estimation unit learning process, the estimation unit 14 learns using data registered in the image database 21 as a learning sample. As a result of this learning, when the feature unit of the image data is input, the estimation unit 14 outputs a disease name estimated from the feature vector. Next, when image data of a diseased part whose disease name is unknown is input to the image processing apparatus 100, image processing that presents image data and a disease name similar to the image data will be described with reference to FIG. This process is started when the user instructs the image processing apparatus 100 to perform image processing.

  First, the image acquisition unit 11 acquires image data of a diseased part whose disease name is unknown input from the image input unit 31 (referred to as “query image data”) (step S301). Step S301 is also called an image acquisition step. Then, the feature vector extraction unit 12 extracts a feature vector of the query image data (Step S302). Step S302 is also referred to as a feature vector extraction step.

  Next, the similarity rank calculation unit 13 calculates the similarity between the feature vector of the query image data and each feature vector in the image database 21 (step S303). Various methods can be appropriately used as a method of calculating the similarity. For example, there is a method in which the norm of two vectors for which similarity is calculated is normalized to 1, and the result of calculating the inner product of the two vectors is used as the similarity. The similarity based on the inner product is a value in the range of 0 to 1, and becomes 1 when the two are most similar (the two vectors match). As another method, there is a method in which a vector norm (a distance between two vectors) of a difference between two vectors is used as a similarity. The similarity based on the norm is 0 or more, and is 0 when the similarity is the most similar (the two vectors match). Note that here, the simple norm refers to the Euclidean norm (the square root of the sum of the squares of each element).

  Then, the similarity rank calculation unit 13 sorts the image data in the image database 21 in descending order of similarity with the query image data calculated in step S303, and sets the rank of each image data in the image database 21. Calculate (step S304). Step S304 is also called a similarity ranking calculation step. For example, when similarity by inner product is adopted as the similarity, sorting is performed in descending order of similarity (inner product), and when similarity by norm is adopted as similarity, sorting is performed in ascending order of similarity (norm). And calculate the ranking.

  Next, the control unit 10 inputs the feature vector of the query image data to the estimation unit 14, and the estimation unit 14 outputs a disease name estimated from the feature vector of the query image data (step S305). Step S305 is also called an estimation step. Then, as a result of the similarity rank calculation in step S304, the rank correction unit 15 matches the disease name associated with the image data having the first similarity degree and the disease name estimated in step S305. It is determined whether or not to perform (step S306). If the disease names match (step S306; Yes), the process proceeds to step S308. In step S306, the rank correction unit 15 functions as a determination unit. Step S306 is also called a determination step.

  If the disease names do not match (step S306; No), in the rank obtained as a result of the sorting in step S304, the rank correction unit 15 sets the rank of the image data associated with the disease name estimated by the estimation unit 14 to N ( For example, the order is corrected by raising 1) (step S307). Step S307 is also referred to as a rank correction step.

  Then, the presentation unit 16 presents the image data and the disease name from the output unit 33 in order from the first place according to the order of each image data (step S308). Step S308 is also referred to as a presentation step. In step S308, the presentation unit 16 presents at least the first-ranked image data or disease name from the output unit 33, but it is possible to appropriately set how many ranks are presented. Then, the control unit 10 ends the image processing.

  FIG. 7 shows how the order is corrected in step S307 when N = 1. In FIG. 7, in the sorting based on the similarity in step S304, the first disease name is a mole, but since the disease name estimated by the estimation unit 14 in step S305 is melanoma, the rank of melanoma is determined in step S307. It is shown that it is one up. As a result, the first-ranked disease name becomes melanoma, and the query image can obtain a diagnosis result that the possibility of melanoma is high.

  A mechanism in which the rank calculated by the similarity rank calculation unit 13 is corrected by the estimation of the disease name by the estimation unit 14 will be described with reference to FIG. Actually, the number of dimensions of the feature vector of the image data is larger than 2, but for the sake of clarity, it is assumed here that the feature vector of the image data is a two-dimensional vector and there are two types of disease names to be identified. To do. In FIG. 8, ◯ and ● indicate the value of the mole feature vector, and ☆ and ★ indicate the value of the melanoma feature vector. Also, ■ indicates the value of the feature vector F1 of the query image. The boundary line L is a boundary line for identifying a mole and a melanoma by the linear SVM included in the estimation unit 14. It was obtained as a result.

  At this time, in FIG. 8, the feature vector most similar to ■ (query image feature vector F1) is ● (mole feature vector F2), and the second most similar feature vector is ★ (melanoma feature vector F3). . That is, as a result of the similarity rank calculation by the similarity rank calculation unit 13, the feature vector that is the first rank is F2 ●, and the disease name is the mole. However, the estimation unit 14 estimates that the name of the disease is melanoma because the black square indicated by F1 is at the lower right of the boundary line L. Since the first disease name by sorting is different from the disease name estimated by the estimation unit 14, the rank of the feature vector corresponding to the disease name estimated by the estimation unit 14 is increased, and the feature of F3 ★ The vector rank is corrected to the first rank. As can be seen from FIG. 8, in the vicinity of the boundary line L, there are some cases where the disease vectors are different even though the feature vectors are similar. Therefore, identification accuracy can be improved by identifying the disease name in consideration of the estimation result by the estimation unit 14 in addition to the similarity.

  In the above example, N, which is a numerical value that the rank correction unit 15 increases the rank, is set to 1. This is because if the estimation unit 14 makes a mistake, the wrong disease name becomes the first. This is to reduce the possibility. When the estimation performance by the estimation unit 14 is high, N may be a value of 2 or more. Further, when the estimation unit 14 can output the reliability of estimation (for example, the reliability is considered to be high if the distance between the feature vector of the query image data and the boundary of the SVM is large), depending on the high reliability. It is also possible to change the value of N dynamically.

  Moreover, in said step S308, although the presentation part 16 presents image data and a disease name in order from the 1st place according to the order of each image data, such presentation is not essential. For example, the presentation unit 16 may present the determination result in step S306 (whether or not the disease names match) from the output unit 33.

Further, as described above, the estimation unit 14 can use not only a linear SVM but also a neural network. However, the use of the linear SVM has the following features.
・ Processing is fast.
A relatively high identification accuracy can be obtained even with a small number of learning samples.
・ Parameter adjustment to improve identification accuracy is relatively easy.

Further, when a neural network is used for the estimation unit 14, the following features are obtained.
-If it learns using a sufficient amount of learning samples, identification accuracy will become higher than SVM.
However, when a neural network is used, there are the following disadvantages.
・ Processing is slow.
・ If the number of learning samples is small, the identification accuracy is low.
-It is difficult to adjust parameters to improve identification accuracy.

  Therefore, according to the number of learning samples, the classifier used in the estimation unit 14 may be SVM or neural network. For example, a neural network can be used if the number of learning samples is greater than or equal to a reference number (for example, 10,000), and SVM can be used if the number is less than the reference number. In this way, even if the number of learning samples is less than the reference number, relatively high discrimination accuracy can be obtained by SVM, and if the number of learning samples is equal to or greater than the reference value, higher discrimination accuracy can be obtained by the neural network.

  In the above embodiment, the similarity rank calculation unit 13 and the estimation unit 14 calculate and estimate the degree of similarity using the feature vector extracted from the image data by the feature vector extraction unit 12. However, the similarity calculation and estimation need not be limited to those using feature vectors. For example, methods such as pattern matching of image data itself, pattern matching of an image divided in a grid pattern, extraction of contour lines from an image, and pattern matching of contour lines may be employed. Also in the case of using a feature vector, the feature vector used by the similarity rank calculation unit 13 and the feature vector used by the estimation unit 14 may be made into feature vectors by different generation methods. In this case, the image database 21 is provided with two feature vector items for each image data, and one is used by the similarity rank calculation unit 13 and the other is used by the estimation unit 14.

  Moreover, in the said embodiment, the image input from the image input part 31 is an image of the affected part of a disease, and the name input from the name input part 32 was the name of the disease. However, it is not necessary to limit to this. For example, the image input from the image input unit 31 may be an image of food (for example, bread), and the name input from the name input unit 32 may be the name of the food (for example, bread). The image input from the image input unit 31 may be an image of a car, and the name input from the name input unit 32 may be the name of the car. The present invention is not limited to these, and the above-described embodiment can be applied to any image as long as the image can be given a name.

  Each function of the image processing apparatus 100 can also be implemented by a computer such as a normal PC (Personal Computer). Specifically, in the above embodiment, the image processing program performed by the image processing apparatus 100 has been described as being stored in advance in the ROM of the storage unit 20. However, the program is stored and distributed on a computer-readable recording medium such as a flexible disk, CD-ROM (Compact Disc Read Only Memory), DVD (Digital Versatile Disc), and MO (Magneto-Optical Disc). A computer capable of realizing each of the functions described above may be configured by reading and installing the program on a computer.

  As mentioned above, although preferable embodiment of this invention was described, this invention is not limited to the specific embodiment which concerns, This invention includes the invention described in the claim, and its equivalent range It is. Hereinafter, the invention described in the scope of claims of the present application will be appended.

(Appendix 1)
An image acquisition unit for acquiring image data;
An image database for storing the image data and a name related to the image data in association with each other;
The similarity indicating the degree of similarity to the image data acquired by the image acquisition unit is obtained for each image data in the image database, and the first rank of the similarity of the image data in the image database is calculated. A similarity ranking calculation unit,
An estimation unit that estimates a name related to the image data acquired by the image acquisition unit;
A determination unit that determines whether a name associated with the first-order image data calculated by the similarity rank calculation unit matches a name estimated by the estimation unit;
A presentation unit that performs a predetermined presentation according to a result determined by the determination unit;
An image processing apparatus comprising:

(Appendix 2)
The similarity rank calculation unit calculates the rank in the image database of the similarity of each image data;
The image processing apparatus according to appendix 1.

(Appendix 3)
When the determination unit determines that the name calculated by the similarity rank calculation unit is different from the name associated with the first image data and the name estimated by the estimation unit, the similarity rank calculation unit Among the image data whose rank is calculated, a rank correction unit that corrects the rank of the image data in which the name associated with the image data matches the name estimated by the estimation unit,
The presenting unit presents at least the first image data in the order of the ranks corrected by the rank correction unit.
The image processing apparatus according to appendix 2.

(Appendix 4)
When the rank calculated by the similarity rank calculation unit is different from the name associated with the first image data and the name estimated by the estimation unit, the rank correction unit calculates the rank In the image data calculated, the rank of the image data in which the name associated with the image data matches the name estimated by the estimation unit is increased by 1.
The image processing apparatus according to appendix 3.

(Appendix 5)
The estimation unit outputs a reliability when estimating the name,
When the rank calculated by the similarity rank calculation unit is different from the name associated with the first image data and the name estimated by the estimation unit, the rank correction unit calculates the rank In the image data calculated, the rank of the image data in which the name associated with the image data matches the name estimated by the estimation unit is increased by a value corresponding to the reliability.
The image processing apparatus according to appendix 3.

(Appendix 6)
A feature vector extraction unit that extracts a feature vector from the image data acquired by the image acquisition unit;
The image database stores the feature vector extracted by the feature vector extraction unit in association with the image data,
The similarity rank calculation unit obtains the similarity using the feature vector extracted by the feature vector extraction unit,
The estimation unit estimates a name related to the image data using the feature vector extracted by the feature vector extraction unit;
The image processing apparatus according to any one of appendices 1 to 5.

(Appendix 7)
The similarity ranking calculation unit obtains the similarity by obtaining a distance between the feature vectors;
The image processing apparatus according to appendix 6.

(Appendix 8)
The estimation unit estimates the name by SVM (Support Vector Machine).
The image processing apparatus according to any one of appendices 1 to 7.

(Appendix 9)
The estimation unit estimates the name by a neural network.
The image processing apparatus according to any one of appendices 1 to 7.

(Appendix 10)
If the number of learning samples used for learning by the estimation unit is greater than or equal to the reference number,
The estimation unit estimates the name by a neural network,
If the number of learning samples used for learning of the estimation unit is less than the reference number,
The estimation unit estimates the name by SVM.
The image processing apparatus according to any one of appendices 1 to 7.

(Appendix 11)
The image data is image data of an affected area of a disease,
The name is the name of the disease,
The image processing apparatus according to any one of appendices 1 to 10.

(Appendix 12)
An image acquisition step for acquiring image data;
A degree of similarity indicating the degree of similarity to the image data acquired in the image acquisition step is obtained for each image data in an image database in which the image data and a name related to the image data are linked and stored, and the image data A similarity rank calculating step of calculating the first rank of the similarity in the image database;
An estimation step of estimating a name related to the image data acquired in the image acquisition step;
A determination step of determining whether or not the name associated with the first rank image data calculated in the similarity rank calculation step matches the name estimated in the estimation step;
A presenting step for performing predetermined presentation according to the result determined in the determining step;
An image processing method including:

(Appendix 13)
On the computer,
An image acquisition step for acquiring image data;
A degree of similarity indicating the degree of similarity to the image data acquired in the image acquisition step is obtained for each image data in an image database in which the image data and a name related to the image data are linked and stored, and the image data A similarity rank calculating step for calculating the first rank of the similarity in the image database;
An estimation step for estimating a name related to the image data acquired in the image acquisition step;
A determination step of determining whether or not the name associated with the first rank image data calculated in the similarity rank calculation step matches the name estimated in the estimation step; and
A presenting step for performing predetermined presentation according to the result determined in the determining step;
A program for running

DESCRIPTION OF SYMBOLS 10 ... Control part, 11 ... Image acquisition part, 12 ... Feature vector extraction part, 13 ... Similarity order calculation part, 14 ... Estimation part, 15 ... Order correction part, 16 ... Presentation part, 20 ... Storage part, 21 ... Image database 31 ... Image input unit, 32 ... Name input unit, 33 ... Output unit, 100 ... Image processing device, F1 ... Feature vector of query image, F2 ... Feature vector of mole, F3 ... Feature vector of melanoma, L ... Boundary line

Claims (13)

An image acquisition unit for acquiring image data;
An image database for storing the image data and a name related to the image data in association with each other;
The similarity indicating the degree of similarity to the image data acquired by the image acquisition unit is obtained for each image data in the image database, and the first rank of the similarity of the image data in the image database is calculated. A similarity ranking calculation unit,
An estimation unit that estimates a name related to the image data acquired by the image acquisition unit;
A determination unit that determines whether a name associated with the first-order image data calculated by the similarity rank calculation unit matches a name estimated by the estimation unit;
A presentation unit that performs a predetermined presentation according to a result determined by the determination unit;
An image processing apparatus comprising: The similarity rank calculation unit calculates the rank in the image database of the similarity of each image data;
The image processing apparatus according to claim 1. When the determination unit determines that the name calculated by the similarity rank calculation unit is different from the name associated with the first image data and the name estimated by the estimation unit, the similarity rank calculation unit Among the image data whose rank is calculated, a rank correction unit that corrects the rank of the image data in which the name associated with the image data matches the name estimated by the estimation unit,
The presenting unit presents at least the first image data in the order of the ranks corrected by the rank correction unit.
The image processing apparatus according to claim 2. When the rank calculated by the similarity rank calculation unit is different from the name associated with the first image data and the name estimated by the estimation unit, the rank correction unit calculates the rank In the image data calculated, the rank of the image data in which the name associated with the image data matches the name estimated by the estimation unit is increased by 1.
The image processing apparatus according to claim 3. The estimation unit outputs a reliability when estimating the name,
When the rank calculated by the similarity rank calculation unit is different from the name associated with the first image data and the name estimated by the estimation unit, the rank correction unit calculates the rank In the image data calculated, the rank of the image data in which the name associated with the image data matches the name estimated by the estimation unit is increased by a value corresponding to the reliability.
The image processing apparatus according to claim 3. A feature vector extraction unit that extracts a feature vector from the image data acquired by the image acquisition unit;
The image database stores the feature vector extracted by the feature vector extraction unit in association with the image data,
The similarity rank calculation unit obtains the similarity using the feature vector extracted by the feature vector extraction unit,
The estimation unit estimates a name related to the image data using the feature vector extracted by the feature vector extraction unit;
The image processing apparatus according to claim 1. The similarity ranking calculation unit obtains the similarity by obtaining a distance between the feature vectors;
The image processing apparatus according to claim 6. The estimation unit estimates the name by SVM (Support Vector Machine).
The image processing apparatus according to claim 1. The estimation unit estimates the name by a neural network.
The image processing apparatus according to claim 1. If the number of learning samples used for learning by the estimation unit is greater than or equal to the reference number,
The estimation unit estimates the name by a neural network,
If the number of learning samples used for learning of the estimation unit is less than the reference number,
The estimation unit estimates the name by SVM.
The image processing apparatus according to claim 1. The image data is image data of an affected area of a disease,
The name is the name of the disease,
The image processing apparatus according to claim 1. An image acquisition step for acquiring image data;
A degree of similarity indicating the degree of similarity to the image data acquired in the image acquisition step is obtained for each image data in an image database in which the image data and a name related to the image data are linked and stored, and the image data A similarity rank calculating step of calculating the first rank of the similarity in the image database;
An estimation step of estimating a name related to the image data acquired in the image acquisition step;
A determination step of determining whether or not the name associated with the first rank image data calculated in the similarity rank calculation step matches the name estimated in the estimation step;
A presenting step for performing predetermined presentation according to the result determined in the determining step;
An image processing method including: On the computer,
An image acquisition step for acquiring image data;
A degree of similarity indicating the degree of similarity to the image data acquired in the image acquisition step is obtained for each image data in an image database in which the image data and a name related to the image data are linked and stored, and the image data A similarity rank calculating step for calculating the first rank of the similarity in the image database;
An estimation step for estimating a name related to the image data acquired in the image acquisition step;
A determination step of determining whether or not the name associated with the first rank image data calculated in the similarity rank calculation step matches the name estimated in the estimation step; and
A presenting step for performing predetermined presentation according to the result determined in the determining step;
A program for running

Images