WO2017110640A1

WO2017110640A1 - Image-processing device, image-processing method, and computer program

Info

Publication number: WO2017110640A1
Application number: PCT/JP2016/087324
Authority: WO
Inventors: 伊藤　直樹
Original assignee: キヤノン株式会社
Priority date: 2015-12-25
Filing date: 2016-12-15
Publication date: 2017-06-29
Also published as: JP2017117335A

Abstract

This image-processing device is characterized by having a classification means for classifying an input image to be processed using a feature value and classification rule acquired from the input image to be processed, a storage means for storing a file that includes the input image to be processed in a storage location determined on the basis of the result of classification by the classification means, an acquisition means for acquiring an operation history for the file stored by the storage means, and a generation means for regenerating the classification rule using the operation history acquired by the acquisition means, the classification means classifying the input image to be processed using the classification rule regenerated by the generation means.

Description

Image processing apparatus, image processing method, and computer program

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

There is a technology that uses a scanner or the like to generate image data of a form generated as a paper medium, identify the type of the form, and perform an OCR (Optical Character Recognition) process. In such a technique, in order to identify a form type and perform OCR processing, it is necessary to create a format definition (form definition) for each form type in advance for all form types. However, creating such a format definition in advance requires time and effort for the user or provider of the form type identification service. Therefore, in order to estimate the form type without creating a format definition in advance, the information processing apparatus accumulates feature data relating to the appearance characteristics of the form for each form type. Then, image data of the processing target form is acquired, feature data related to the processing target form is acquired, and feature data that matches or resembles the feature data is specified from the stored feature data (matching processing is performed) ) To estimate the form type. If the form type is not estimated, feature data is added (see).

Patent No. 5670787

When using a method for recognizing a document type by matching processing as in the conventional technology, the document recognition result is presented to the user every time, and the user is made to determine whether or not the recognition result is correct. Then, the determination result by the user is transmitted to the document recognition system.

Thus, in order to perform document recognition processing, it is necessary to present to the user whether or not the classification result matches the classified document and to receive an instruction of the determination result from the user. Therefore, the document recognition is not completed until the user instructs the determination result.

The present invention solves the above-described problem, and provides an image processing apparatus capable of generating a document classification rule, which is used when performing a document identification process by a natural operation without requiring an explicit instruction from a user. For the purpose of provision.

The image processing apparatus determines a classification unit that classifies the input image to be processed using a feature amount and a classification rule acquired from the input image to be processed, and a result classified by the classification unit. A storage unit for storing a file including the input image to be processed in the storage location, an acquisition unit for acquiring an operation history for the file stored by the storage unit, and an operation history acquired by the acquisition unit. Generation means for regenerating the classification rule, wherein the classification means classifies the input image to be processed using the classification rule regenerated by the generation means. .

According to the image processing apparatus of the present invention, it is possible to generate a document classification rule that is used when performing a document identification process by performing a natural operation unintentionally by a user.

1 is an overview diagram showing a system configuration in Embodiment 1. FIG. 1 is a block diagram illustrating a schematic configuration of an MFP according to a first embodiment. 1 is a block diagram illustrating a schematic configuration of an MFP according to a first embodiment. 1 is a block diagram illustrating a schematic configuration of a server in Embodiment 1. FIG. 3 is a flowchart in the first embodiment. FIG. 6 is a diagram illustrating calculation of an image feature amount based on gradient information in the first embodiment. FIG. 6 is a diagram for explaining extraction of a patch image for calculating an image feature amount according to the first exemplary embodiment. It is a figure explaining the increase process of the learning data in Example 1. FIG. It is a figure explaining the increase process of the learning data in Example 1. FIG. It is a figure explaining the increase process of the learning data in Example 1. FIG. It is a figure explaining learning data in Example 1, and explaining classification processing. 10 is a flowchart in the second embodiment.

Example 1
FIG. 1 is a diagram illustrating a system configuration of the first embodiment.

The system according to the first embodiment includes an MFP 101 serving as an image input device and a server 102.

The MFP 103 is connected to the LAN 103. The LAN 103 is connected to the Internet 104 and is also connected to the server 102 that provides the service via the Internet 104. The MFP 101 and the server 102 are connected to each other via the LAN 103 and transmit / receive image data and various types of information. In this example, the MFP 101 is connected to the LAN 103, but this is not a limitation. The MFP 101 only needs to be connectable to the server 102. The server 102 is an information processing apparatus that generates learning data used for learning classification rules for images input from the image input apparatus.

Here, the “classification rule” is a combination of a feature amount acquired from an image and a classifier for classifying (identifying) the image. In order to classify an image having a certain feature into category 1 and an image having another feature into category 2, a rule (rule) is provided in advance.

In addition, “generate learning data” is to generate a document classification rule used when performing document identification processing.

Specifically, the server 102 generates learning data in which the rotation direction of the image input from the image input device is unified, and constructs an image classification rule based on the generated learning data.

2A and 2B are diagrams illustrating a configuration example of the MFP 101. FIG.

As shown in FIG. 2A, the MFP 101 includes a controller 20 to an operation unit 207.

The device control unit 200 exchanges data with the outside of the MFP 101 and the outside via the network I / F 206 and accepts an operation from the operation unit 207.

The image reading unit 201 reads a document and outputs image data.

The image processing unit 202 converts print information including image data input from the image reading unit 201 or the outside into intermediate information (hereinafter referred to as “object”), and stores it in the object buffer of the storage unit 203. Objects have text, graphic, and image attributes. Further, bitmap data is generated based on the buffered object and stored in the buffer of the storage unit 203. At that time, pseudo halftone processing such as color conversion processing, density adjustment processing, toner total amount control processing, video count processing, printer gamma correction processing, and dithering is performed.

The storage unit 203 includes a ROM, a RAM, a hard disk (HDD), and the like. The ROM stores various control programs and image processing programs executed by the CPU 204. The RAM is used as a reference area or work area in which the CPU 204 stores data and various types of information. The RAM and HDD are used for the object buffer and the like. Image data is stored on the RAM and HDD, the pages are sorted, the originals that are sorted over a plurality of pages are stored, and a plurality of copies are printed out.

The image output unit 205 forms and outputs a color image on a recording medium such as recording paper. A network I / F unit 206 connects the MFP 101 to the LAN 103 and transmits / receives various information to / from the Internet 104 and other devices.

The operation unit 207 includes a touch panel and operation buttons, receives an operation from the user, and transmits information on the operation to the device control unit 200.

FIG. 2B shows the appearance of the MFP 101. The image reading unit 201 has a plurality of CCDs. If the sensitivity of each CCD is different, it is recognized that each pixel has a different density even if the density of each pixel on the document is the same. Therefore, the image reading unit first performs exposure scanning on a white plate (uniformly white plate), converts the amount of reflected light obtained by the exposure scanning into an electrical signal, and outputs it to the controller. The shading correction unit in the image processing 202 recognizes the difference in sensitivity of each CCD based on the electrical signal obtained from each CCD. Then, using the recognized difference in sensitivity, the value of the electric signal obtained by scanning the image on the document is corrected. Further, when the shading correction unit receives gain adjustment information from the CPU 204 in the controller, the shading correction unit performs gain adjustment according to the information. The gain adjustment is used to adjust how the value of the electrical signal obtained by exposing and scanning the document is assigned to the luminance signal value of 0 to 255. By this gain adjustment, the value of the electrical signal obtained by exposing and scanning the document can be converted into a high luminance signal value or converted into a low luminance signal value. That is, the dynamic range of the read signal can be adjusted by this gain adjustment. Next, a configuration for scanning the image on the document will be described.

The image reading unit 201 converts the image information into an electrical signal by inputting the reflected light obtained by exposing and scanning the image on the document to the CCD. Further, the electrical signal is converted into a luminance signal composed of R, G, and B colors, and the luminance signal is output as an image to the controller 20.

Note that the document is set on the tray 212 of the document feeder 211. When the user gives an instruction to start reading from the operation unit 207, a document reading instruction is given from the controller 20 to the image reading unit 201. Upon receiving this instruction, the image reading unit 201 feeds originals one by one from the tray 212 of the original feeder 211, and performs an original reading operation. The document reading method is not an automatic feeding method using the document feeder 211, but may be a method of scanning the document by placing the document on a glass surface (not shown) and moving the exposure unit.

The image output unit 205 is an image forming device that forms an image received from the controller 20 on a sheet. In this embodiment, the image forming method is an electrophotographic method using a photosensitive drum or a photosensitive belt, but the present embodiment is not limited to this. For example, the present invention can also be applied to an ink jet system that prints on paper by ejecting ink from a micro nozzle array. The image output unit 205 is provided with a plurality of

paper cassettes

213, 214, and 215 that can select different paper sizes or different paper orientations. The printed paper is discharged to the paper discharge tray 216.

FIG. 3 is a diagram illustrating a configuration example of the server.

The server 102 includes a CPU 301 to a data bus 306.

CPU 301 reads a control program stored in ROM 303 and executes various control processes. The RAM 302 is used as a temporary storage area such as a main memory or work area for the CPU 301. A network I / F unit 304 connects the server 102 to the Internet 104 and transmits / receives various information to / from other devices. The HDD 305 stores image data, feature amount data, and various programs.

3, the image data from the MFP 101 received via the network I / F unit 304 is sent and received by the CPU 301, RAM 302, and ROM 303 via the data bus 306. When the CPU 301 executes an image processing program stored in the ROM 303 or the HDD 305, image processing for the image data is realized.

<Detailed description of this embodiment using flowchart>
In FIG. 4, a classification process is performed on an image obtained by scanning a document, and a classification destination (storage destination) is determined according to the result. And it is a flowchart explaining the process which acquires the operation log (operation history) with respect to an image (file containing this image), and re-learns (regenerates) the image classification rule.

The processing shown in this flowchart is executed by the MFP 101 and the server 102. Of the processes shown in the flowchart of FIG. 4, the processes executed by the MFP 101 are realized by the CPU 204 loading and executing a processing program stored in the storage unit 203. 4 is realized by the CPU 301 loading the processing program stored in the HDD 305 into the RAM 302 and executing the processing.

In step S 401, when the MFP 101 receives a user instruction from the operation unit 207, the MFP 101 feeds documents one by one from the tray 212 of the document feeder 211, and scans the document by the image reading unit 201.

In step S402, the CPU 204 executes a process for acquiring the feature amount of the input image obtained by scanning the document. The feature amount of the image will be described later with reference to FIGS. 6, 7A, 7B, and 7C.

In step S403, the CPU 204 stores the feature amount acquired in step S402 in the storage unit 203.

In step S404, the CPU 204 performs a classification process according to the feature amount stored in the storage unit 203 in step S403 and the classification rule created by learning in advance. As a result of the classification process, for example, an ID number of a previously learned image (learned image) is associated with a feature amount acquired from the image, and this feature is obtained for an image having a certain feature amount. An ID number corresponding to the quantity is output. Therefore, when the input image is classified as the learning image ID1 as a result of the classification process, the corresponding image ID “1” is output. If the input image cannot be classified as a learned image, an image ID indicating an unknown form (unknown image) is output. Here, with regard to the classification process, a known technique such as classification using machine learning can be applied.

In step S405, the CPU 204 determines whether the input image in step S404 is classified as a learned image or an unlearned unknown image as a result of the classification. Here, using the image ID of the classification result described in step S404, it is determined whether the image is classified as a learned image or an unknown image. It is possible to apply a method that can determine whether the image is classified as a learned image or an unknown image by other methods.

In step S <b> 406, the CPU 201 determines whether or not the transmission destination for saving the input image is a transmission destination that can acquire information on the operation of the file including the input image from the MFP 101. For example, the CPU 204 determines whether or not the transmission destination is beyond the firewall via the network I / F unit 206. If there is an input image transmission destination beyond the firewall, it is determined that there is no access right to the input image storage destination. Here, whether or not there is an access right indicates whether or not an operation log performed on a file in a save destination folder can be acquired.

Here, it is possible to apply a publicly known method for detecting whether or not a file operation log can be acquired for a storage destination folder.

In step S407, the CPU 204 performs an input image storage process when it is determined in step S406 that the transmission destination has the access right. As an example to be described here, it is determined that the input image storage destination has access right, and the classification processing could not be performed at the location determined to have this access right (the image was unknown). Saved to a folder for saving images. As a specific example, the input image is stored in a folder named “unknown” (hereinafter referred to as an unknown folder).

In step S408, the MFP 101 acquires an operation log for the file saved in step S407. Here, the operation log is information about operations performed to move folders to files, operations performed to change file names, and operations performed to delete files. It is. As a specific example, an operation log of an operation in which an image stored in an unknown folder is moved to which folder is acquired.

In step S409, the CPU 204 generates feedback information from the operation log acquired in step S408. When generating feedback information, an operation log indicating where an image stored in an unknown folder is moved to is used to newly generate an image stored in the unknown folder. As a specific example of assigning an image ID, for example, using an operation log indicating that a first image stored in an unknown folder has been moved to a folder called Document 1, a document is applied to the first image. An image ID for entering 1 is generated and assigned. This assigned information is used as feedback information.

In step S410, the CPU 204 performs a relearning process from the feedback information generated in step S409. In the relearning process, a threshold value update process used in performing the classification process is performed using the feature amount held in step S403 and the feedback information generated in step S409. For example, image ID information is acquired from the feedback information, and a threshold value calculation process is performed so that the image feature amount is classified into the image ID. Here, the updated threshold value is reflected in the classification process, and the classifier used for classification is updated (reproducing the document classification rule used when performing the document identification process).

In step S411, if it is determined in step S406 that there is no access right for the storage destination of the input image, the CPU 204 stores the file in a storage destination having a separate access right. For example, the data is stored in the storage unit 203 of the MFP 101.

In step S412, the MFP 101 acquires an operation log for the file saved in step S411. Since the acquisition of the operation log here is the same as that described in step S408, description thereof is omitted.

On the other hand, if it is determined in step S405 that the image is a known image, the process proceeds to step S413. In step S413, the CPU 204 performs storage processing in accordance with a rule designated in advance.

<Details of image features used for image classification>
Details of the image feature amount used in the image classification process will be described with reference to FIGS. 5 and 6. FIG. 5 is a diagram for explaining a method for calculating an image feature amount based on gradient information.

As shown in FIG. 5, the gradient strength and gradient direction calculated for each pixel in the patch image are used. Specifically, the CPU 301 obtains the gradient strength and gradient direction from the edge information in the vertical direction and the horizontal direction for all pixels in the patch image. The CPU 301 uses the gradient information to calculate 9-dimensional (9) feature amounts from one patch, as shown in FIG. First, for each pixel, a pixel having a gradient strength equal to or greater than a certain value is defined as an edge pixel, and a pixel having a gradient intensity smaller than the certain value is defined as a non-edge pixel. The gradient direction is quantized into 8 directions from the edge pixel group, and the gradient intensity integrated value / patch pixel number for each direction is calculated, and combined with the non-edge pixel number / patch pixel number, a 9-dimensional feature from one patch image Calculate the amount. As described above, by using the edge pixels and the non-edge pixels, it is possible to express not only ruled line and character information but also a margin part which is a large feature of the document image. The description so far is the description of the feature amount in one patch image, but actually, a large number of feature amounts are used by cutting out and using a plurality of patch images.

FIG. 6 is a diagram for explaining extraction of a patch image.

First, the CPU 301 deletes image edges (margins) where noise is likely to appear, and creates images with a plurality of resolutions. The reason for preparing images with a plurality of resolutions is that the structure of the edge changes for each resolution. Then, the CPU 301 calculates a feature amount in consideration of the patch image position by cutting out patch images of a plurality of sizes from the respective resolution images. For example, assume that a feature amount is extracted from an image scanned at 300 dpi. First, the CPU 301 creates two types of images obtained by reducing the scanned image to ¼ size and ８ size. The CPU 401 shifts the 1/4 size patch image by 1/5 from each reduced resolution image by 5 × 5 = 25 sheets, and shifts the 1/8 size patch image by 1/10 by 10 × 10. By cutting out 100 sheets, a total of 250 patches are cut out. In such a setting, since a 9-dimensional edge feature amount is calculated from each patch, it is possible to calculate a 9 × 250 = 2250-dimensional feature amount from one image.

Note that the parameters relating to the image resolution, patch size, and patch cut-out position are not limited to the numbers described above. Further, in order to use document color information as an image feature amount to be acquired, a color histogram, color dispersion, or the like may be used as the image feature amount.

<Details of learning data increase processing>
In the image direction unification process, a learning data increasing process for increasing learning data when generating a direction discriminator using machine learning will be described. In the present embodiment, a deformed image is obtained by performing a deformation process on the image by simulation, and this is increased as learning data.

7A, 7B, and 7C are diagrams illustrating shift processing, rotation processing, and enlargement / reduction processing that are deformation processing. These geometric deformation processes are realized using a projective transformation matrix. FIG. 7A shows the shift process. In the shift process, eight patterns of deformed images are obtained by moving the images in parallel by a fixed amount in the vertical and horizontal directions or in the upper left, upper right, lower left and lower right. FIG. 7B shows the rotation process. In the rotation process, two patterns of deformed images are obtained by rotating clockwise and counterclockwise by a certain amount. FIG. 7C shows the enlargement / reduction processing. In the enlargement / reduction process, two patterns of deformed images are obtained by enlarging or reducing the image by a predetermined magnification. In each transformation process, the input image and the output image have the same size. The outside image area that protrudes outside the image area of the output image after projective transformation is discarded. In addition, a missing region in which no projection source exists in the output image is complemented by sequentially copying pixel values of non-missing pixels. The handling of this missing area is not limited to complementing by the method described above. For example, another complementing method that replaces the background pixel estimated from the input image may be used, or a method of adding flag information that the missing pixel is a missing pixel without performing the complementing may be used.

In the learning data increase process, by combining the patterns that are not deformed into the shift process, the rotation process, and the enlargement / reduction process, it is possible to obtain deformed images corresponding to the number of combinations from one image data. Specifically, in addition to the case where there is no correction to the above-described deformation process patterns, there are 9 shift processes, 3 rotation processes, and 3 enlargement / reduction processes. 9 × 3 = 81 patterns of deformed images are generated to increase learning data. Note that the number of patterns of each deformation process is not limited to the above-described numbers.

<Details of machine learning to be used>
Next, a machine learning method used for generating a classifier that classifies images in this embodiment will be described. In the present embodiment, a known technique called Real AdaBoost is used as a machine learning technique. Real AdaBoost is a method capable of selecting a feature amount suitable for classification of a given learning data set from a large amount of feature amounts and combining the feature amounts to configure a classifier. If a large amount of feature amount is used at the time of image classification, performance deteriorates due to the calculation load of the feature amount. Thus, it is a great advantage of Real AdaBoost that a classifier can be configured by selecting feature quantities suitable for classification and using only some of the feature quantities. However, Real AdaBoost is a two-class classifier and classifies data with two types of labels. That is, as it is, it cannot be used for classification of three or more types of images. Therefore, a known method called OVA (One-Versus-All) that expands the two-class classifier to a multi-class classifier is used. OVA creates classifiers for classifying one class (target class) and other classes by the number of classes, and uses the output of each classifier as the reliability of the target class. At the time of classification, data to be classified is input to all classifiers, and the class having the highest reliability is set as the classification destination.

FIG. 8 is a diagram for explaining an example of machine learning using learning data.

In this example, it is assumed that image feature amounts corresponding to each of three classes of images (image A, image B, and image C) are prepared as learning data. In order to classify these three classes, OVA prepares three types of classifiers. The three types of classifiers are an image A discriminator for discriminating between image A and other images, an image B discriminator for discriminating between image B and other images, and an image B discriminating device for discriminating between image C and other images. An image C discriminator.

The image A discriminator outputs a large output value (confidence) when the image A is input, and outputs a small output value (confidence) when other images are input. The same applies to the image B classifier and the image C classifier. When actual classification is performed, input document images are input to three types of classifiers, and output values are compared to determine which image. For example, when the output of the image B discriminator is maximum, it is determined that the input image is the image B.

The learning of the multi-class classifier using Real AdaBoost and OVA described with reference to FIG. 8 and the document image classification using the multi-class classifier are executed by the CPU 301. Note that the machine learning technique that can be used in the present embodiment is not limited to the technique described above. You may utilize well-known methods, such as Suppprot Vector Machine and Random Forest. If the feature selection framework is not included in the machine learning method and you want to improve the classification speed during classification, select a known feature value such as feature value selection using principal component analysis or discriminant analysis. Do. When the device learning method is a two-class classifier, a known method such as All-Versus-All (AVA) or Error- Correcting Output-Coding (ECOC) other than OVA may be used.

As described above, according to the first embodiment, an operation log for a user's file is acquired for an input image whose classification destination (storage destination) is unknown, and document identification processing is performed by using the acquired operation log. It is possible to generate (relearn) a document classification rule to be used when performing. That is, the classifier used for document classification is updated.

This makes it possible to classify images even for images that do not have classification rules (no learning) without the user explicitly instructing re-learning.

(Example 2)
In the first embodiment, the operation for the image stored in the unknown folder explained the folder moving process.

In Example 2, it is assumed that a deletion operation is performed on an image stored in an unknown folder. Below, only the part which has a difference with Example 1 is demonstrated.

<Detailed description of this embodiment using flowchart>
FIG. 9 is a flowchart for describing processing for detecting whether or not a file including an input image has been deleted from an operation log for an unclassified input image and generating feedback information in response to the deletion. Of the processes shown in FIG. 9, the process executed by the MFP 101 is realized by the CPU 204 loading and executing a processing program stored in the storage unit 203. Steps S401 to S413 are omitted since they have been described in the first embodiment.

In step S901, the CPU 204 determines whether the file has been deleted from the log acquired in step S408 or step S412. Detection of whether or not a file has been deleted can be obtained by a known method.

In step S902, when the CPU 204 determines that the operation for deleting the file has been performed in step S901, the CPU 204 generates a folder for deletion and stores the image therein. Alternatively, the operation unit 207 is displayed for a user instruction. Here, what is displayed on the operation unit 207 is to cause the user to select whether to create a folder for deletion and move the file there, or delete the file without generating the folder for deletion.

In step S903, the CPU 204 determines whether the user has input to the operation unit 207 when generating a deletion folder or not generating a deletion folder.

In step S904, the CPU 204 generates a deletion folder when instructed to generate a deletion folder in step S903, and stores a file including an image to be deleted in the folder. Then, an image ID is assigned to the file so that the file including the image is classified into the deletion folder.

In step S409, an image ID for classifying the processing target image into a folder for deletion is input.

In step S903, if it is determined that the deletion folder is not generated, the CPU 204 deletes the file, and ends the process without generating feedback information.

According to the second embodiment, an operation log in which an image in an unknown folder is deleted is acquired. For example, when an image stored in an unknown folder is deleted, it is considered that the image is an unnecessary image from the operation log. Therefore, an image ID that is unnecessary or classified as a trash can is assigned to the input image. As a result, it is possible to avoid unnecessary images from being classified into unknown folders many times, so that it is possible to reduce troublesome operations such as the user deleting images classified into unknown folders every time.

(Other examples)
The present invention is also realized by executing the following processing. That is, software (computer program) that realizes the functions of the above-described embodiments is supplied to a system or apparatus via a network or various storage media, and the computer of the system or apparatus (or CPU, MPU, etc.) reads the program. To be executed.

The present invention is not limited to the above embodiment, and various changes and modifications can be made without departing from the spirit and scope of the present invention. Therefore, in order to make the scope of the present invention public, the following claims are attached.

This application claims priority based on Japanese Patent Application No. 2015-254377 filed on December 25, 2015, the entire contents of which are incorporated herein by reference.

Claims

Classification means for classifying the input image to be processed using a feature amount and a classification rule acquired from the input image to be processed;
Storage means for storing a file containing the input image to be processed in a storage location determined based on the result of classification by the classification means;
Obtaining means for obtaining an operation history for the file stored by the storage means;
Generating means for regenerating the classification rule using the operation history acquired by the acquisition means,
The image processing apparatus according to claim 1, wherein the classification unit classifies the input image to be processed using the classification rule regenerated by the generation unit.
2. The image processing apparatus according to claim 1, wherein the classification unit classifies the input image to be processed using a classification rule regenerated by reflecting the operation history.
The image processing apparatus according to claim 1, wherein the classification rule is a combination of a feature amount acquired from an image and a classifier for classifying an input image to be processed.
2. The image processing apparatus according to claim 1, wherein the operation history is information relating to folder movement, file name change, or file deletion for a file including an input image to be processed.
When the storage means does not have access right to the storage destination storing the file including the input image to be processed, the file is stored in a storage destination different from the storage destination and having access right. The image processing apparatus according to claim 1.
The image processing apparatus according to claim 1, further comprising display control means for performing display for receiving an instruction on a screen when the operation history is a specific operation.
A classification step for classifying the input image to be processed using the feature amount and the classification rule acquired from the input image to be processed;
A storage step of storing a file including the input image to be processed in a storage destination determined based on the result of the classification in the classification step;
An acquisition step of acquiring an operation history for the file stored in the storage step;
A generation step of regenerating the classification rule using the operation history acquired in the acquisition step;
An image processing method, wherein classification is performed on an input image to be processed using the regenerated classification rule.
Computer
A classification step for classifying the input image to be processed using the feature amount and the classification rule acquired from the input image to be processed;
A storage step of storing a file including the input image to be processed in a storage destination determined based on the result of the classification in the classification step;
An acquisition step of acquiring an operation history for the file stored in the storage step;
And regenerating the classification rule using the operation history acquired in the acquisition step,
A program for causing an input image to be processed to be classified using the regenerated classification rule as an image processing method.