CN111340056A

CN111340056A - Information processing method and information processing apparatus

Info

Publication number: CN111340056A
Application number: CN201811553508.9A
Authority: CN
Inventors: 李斐; 田虎
Original assignee: Fujitsu Ltd
Current assignee: Fujitsu Ltd
Priority date: 2018-12-18
Filing date: 2018-12-18
Publication date: 2020-06-26
Anticipated expiration: 2038-12-18
Also published as: CN111340056B

Abstract

Provided are an information processing method and an information processing apparatus. The information processing apparatus can be used to detect an anomaly in an image file containing a repetitive pattern, and includes a processor configured to: encoding a current region of an image file serving as a detection target by using a pre-trained self-encoder to obtain a hidden variable of the current region; acquiring similar areas of a current area from an image file, and encoding each similar area by using an autoencoder to acquire hidden variables of each similar area; modifying the hidden variable of the current area based on the acquired hidden variable of the similar area; decoding the modified hidden variable by using an auto-encoder to obtain a reconstructed region of the current region; and comparing the current region with the reconstructed region, and judging whether the current region is abnormal or not based on the comparison result.

Description

Information processing method and information processing apparatus

Technical Field

The present disclosure relates generally to the field of information processing, and more particularly, to an information processing method for detecting an abnormality in an image file including a repetitive pattern and an information processing apparatus capable of implementing the same.

Background

In different technical fields or production industries, it is often involved in the detection and processing of images comprising repetitive patterns. As examples, fabrics having a repetitive texture or design (such as carpets, curtains, clothes, etc.), prints having a repetitive pattern (such as wallpaper, decorative pictures, etc.), even photographs obtained by taking items including a repetitive pattern, etc. may be regarded as images including a repetitive pattern.

In the processing relating to an image including a repetitive pattern, it is generally necessary to detect whether the image has an abnormality or defect. For example, it is necessary to detect whether a pattern that should be periodically repeated originally as a repetitive pattern is not repeated at a predetermined rule. Taking a fabric with a repeating texture as an example, quality checks for defects and the like that do not belong to a repetition according to a predetermined rule play an important role in the textile industry. Traditionally, such anomaly or defect detection is accomplished by manual visual inspection. However, since many defects on the surface of the fabric are small and difficult to identify, manual-based operations are not only time consuming and labor intensive, but the results are unreliable. In addition, although automatic fabric defect detection has been noted and applied in recent years with the development of computer vision, the texture of fabrics has become more and more complex and varied at the same time, thus also making the task of fabric quality inspection more and more difficult. Similar problems exist for other types of images that include repeating patterns.

Accordingly, there is a need to provide a method that can accurately and efficiently perform automatic anomaly detection on an image including a repetitive pattern.

Disclosure of Invention

The following presents a simplified summary of the invention in order to provide a basic understanding of some aspects of the invention. It should be understood that this summary is not an exhaustive overview of the invention. It is not intended to determine the key or critical elements of the present invention, nor is it intended to limit the scope of the present invention. Its sole purpose is to present some concepts in a simplified form as a prelude to the more detailed description that is discussed later.

In view of the need for accurate and efficient automatic anomaly detection for images including repetitive patterns, it is an object of the present invention to provide an information processing method for automatically detecting anomalies in image files including repetitive patterns, and an information processing apparatus capable of implementing the above-described information processing method, which is capable of achieving accurate and efficient automatic anomaly detection.

According to a first aspect of the present disclosure, there is provided an information processing method for detecting an abnormality in an image file containing a repetitive pattern, comprising: encoding a current region of an image file serving as a detection target by using a pre-trained self-encoder to obtain a hidden variable of the current region; acquiring similar areas of a current area from an image file, and encoding each similar area by using an autoencoder to acquire hidden variables of each similar area; modifying the hidden variable of the current area based on the acquired hidden variable of the similar area; decoding the modified hidden variable by using an auto-encoder to obtain a reconstructed region of the current region; and comparing the current region with the reconstructed region, and judging whether the current region is abnormal or not based on the comparison result.

In accordance with another aspect of the present disclosure, there is provided an information processing apparatus which can be used to detect an abnormality in an image file containing a repetitive pattern, and includes a processor configured to: encoding a current region of an image file serving as a detection target by using a pre-trained self-encoder to obtain a hidden variable of the current region; acquiring similar areas of a current area from an image file, and encoding each similar area by using an autoencoder to acquire hidden variables of each similar area; modifying the hidden variable of the current area based on the acquired hidden variable of the similar area; decoding the modified hidden variable by using an auto-encoder to obtain a reconstructed region of the current region; and comparing the current region with the reconstructed region, and judging whether the current region is abnormal or not based on the comparison result.

According to other aspects of the present disclosure, there is also provided a program causing a computer to implement the information processing method as described above.

According to yet another aspect of the present disclosure, there is also provided a corresponding storage medium storing machine-readable instruction code, which, when read and executed by a machine, is capable of causing the machine to perform the information processing method as described above.

The foregoing, in accordance with various aspects of embodiments of the present disclosure, can result in at least the following benefits: with the information processing method and the information processing apparatus provided by the present disclosure, for an image area to be processed in which an abnormality exists, a reconstruction area that substantially does not include an abnormality (or a reconstruction area that is little affected by the abnormality) can be reconstructed with the self-encoder based on the similarity between the areas in the image file having the repetitive pattern, and the abnormality in the image area to be processed can be detected efficiently and accurately based on such a reconstruction area.

These and other advantages of the present disclosure will become more apparent from the following detailed description of the preferred embodiments of the present disclosure when taken in conjunction with the accompanying drawings.

Drawings

The disclosure may be better understood by reference to the following description taken in conjunction with the accompanying drawings, in which like or similar reference numerals identify like or similar parts throughout the figures. The accompanying drawings, which are incorporated in and form a part of this specification, illustrate preferred embodiments of the present disclosure and, together with the detailed description, serve to explain the principles and advantages of the disclosure. Wherein:

fig. 1 is a flowchart schematically showing one example flow of an information processing method according to an embodiment of the present disclosure.

Fig. 2 is a flowchart schematically showing one example process of a hidden variable modification step in the information processing method of fig. 1.

Fig. 3 is a flowchart schematically showing another exemplary process of a hidden variable modification step in the information processing method of fig. 1.

Fig. 4 is a flowchart schematically showing a preferred example flow of an information processing method according to an embodiment of the present disclosure, in which an example process of the hidden variable modification step shown in fig. 3 is adopted.

Fig. 5 is a flowchart schematically showing another preferred example flow of an information processing method according to an embodiment of the present disclosure.

Fig. 6 is a flowchart schematically showing still another preferred example flow of an information processing method according to an embodiment of the present disclosure.

Fig. 7 is a flowchart schematically showing still another preferred example flow of an information processing method according to an embodiment of the present disclosure.

Fig. 8 is a schematic block diagram schematically showing one example structure of an information processing apparatus according to an embodiment of the present disclosure.

Fig. 9 is a schematic block diagram schematically showing one example structure of a hidden variable modification unit in the information processing apparatus of fig. 8.

Fig. 10 is a schematic block diagram schematically showing another exemplary structure of a hidden variable modification unit in the information processing apparatus of fig. 8.

Fig. 11 is a schematic block diagram schematically showing one preferred example structure of an information processing apparatus according to an embodiment of the present disclosure.

Fig. 12 is a block diagram showing one possible hardware configuration that can be used to implement the information processing method and apparatus according to the embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present invention will be described hereinafter with reference to the accompanying drawings. In the interest of clarity and conciseness, not all features of an actual implementation are described in the specification. It will of course be appreciated that in the development of any such actual embodiment, numerous implementation-specific decisions must be made to achieve the developers' specific goals, such as compliance with system-related and business-related constraints, which will vary from one implementation to another. Moreover, it will be appreciated that such a development effort might be complex and time-consuming, but would nevertheless be a routine undertaking for those of ordinary skill in the art having the benefit of this disclosure.

It should be noted that, in order to avoid obscuring the present invention with unnecessary details, only the device structures and/or processing steps closely related to the solution according to the present invention are shown in the drawings, and other details not so relevant to the present invention are omitted.

According to a first aspect of the present disclosure, an information processing method is provided. Fig. 1 is a flowchart schematically showing one example flow of an information processing method according to an embodiment of the present disclosure.

The information processing method 100 shown in fig. 1 may be used to detect anomalies in image files that contain repetitive patterns. As shown in fig. 1, the information processing method 100 may include: a current region encoding step S101, encoding a current region of an image file serving as a detection target by using a pre-trained self-encoder to obtain a hidden variable of the current region; a similar region encoding step S103 of acquiring a similar region of the current region from the image file, and encoding each similar region using an auto-encoder to acquire a hidden variable of each similar region; a hidden variable modification step S105 of modifying a hidden variable of the current region based on the acquired hidden variable of the similar region; a hidden variable decoding step S107, decoding the modified hidden variable by using an auto-encoder to obtain a reconstructed region of the current region; and an abnormality determination step S109 of comparing the current region with the reconstructed region and determining whether there is an abnormality in the current region based on the comparison result.

As an example, the self-encoder used in the information processing method of the present embodiment may be a self-encoder obtained by training using training image data that does not include an anomaly or a defect in the same manner as in the related art.

For the purpose of facilitating the following description, a brief description is given here of the self-encoder. The auto-encoder is an efficient deep neural network that learns data characterizations in an unsupervised manner, and comprises an encoder part and a decoder part, wherein the encoder part can be represented as an encoding function f () for encoding an input signal into hidden variables, and the decoder part can be represented as a decoding function g () for re-decoding the hidden variables obtained by the encoding into a new signal. By training in advance based on a training image or training image block that does not contain a defect, the signal obtained through the above-described encoding and decoding process by the self-encoder is generally good for reconstructing the input signal, and thus the signal may also be referred to as a reconstructed signal. When the detection processing is performed using the self-encoder obtained by such training, an image area containing no defect in the image to be processed can be reconstructed well, and an image area containing a defect cannot be reconstructed, so that the presence of a defect can be detected.

The information processing method of the present embodiment may employ, for example, a self-encoder obtained by training in the above-described manner. More specifically, as an example, it is possible to input training image blocks collected in advance (i.e., image blocks without defects) as training data to an auto-encoder composed of an encoding network (encoder portion) and a decoding network (decoder portion), construct a loss function based on a difference between the input training data and reconstructed data output by the decoding network, and learn parameters of the encoding network and the decoding network by minimizing the loss function, i.e., obtain specific forms and/or parameters of the above-described encoding function f () and decoding function g (). Those skilled in the art can use any suitable existing network model to construct the above-mentioned self-encoder, and can use any suitable existing training mode to learn the parameters therein, which is not described herein again.

The inventors found that the above-described self-encoder is effective for a general defect detection task, but when such a self-encoder is applied to an abnormality detection task such as a fabric quality inspection, since an abnormality such as a defect or the like in an image to be processed may be very minute, an image region where the abnormality exists is very similar to an image region where no abnormality exists, so that many image regions where the abnormality exists can also be well reconstructed by the self-encoder, thereby failing to accurately detect the abnormality in the image to be processed.

In addition, the inventors have made an insight into the fact that most areas in an image (or image file) including a repetitive pattern, such as a fabric, are normal. Therefore, in the information processing method of the present embodiment, when processing the current region where there may be an abnormality, the hidden variable of the current region is modified by using the hidden variable of the similar region that is very likely to be normal (i.e., not including the abnormality) in the image file, so that, for example, a reconstructed region that is similar to the current region and substantially does not include the abnormality (or a reconstructed region that is less affected by the abnormality) can be reconstructed. Based on the result of comparing such a reconstructed region with the current region, it is possible to accurately detect whether there is an abnormality in the current region, and thus it is particularly advantageous for a case where an abnormality such as a stain or the like in the image to be processed is very slight.

The respective steps of the information processing method of the present embodiment will be described below with reference to one specific example.

In the present example, a previously obtained self-encoder is represented by a coding function f () and a decoding function g (), in which the form and parameters of each of the coding function f () and the decoding function g () are previously obtained by training and are not changed during a test or detection process on a target image file. Further, a current area to be processed in the image file is denoted by x.

In this example, for the current region x, first, a current region encoding step S101 of the information processing method 100 is performed, that is, the current region x is encoded by an encoding function f (), and a hidden variable h ═ f (x) of the current region is obtained.

Next, a similar region encoding step S103 is performed, wherein n image regions x similar to the current region x can be found in the whole image of the image file₁，x₂，…，x_n(e.g., n image regions having a similarity higher than a predetermined threshold with respect to the current region x), and a hidden variable h of each similar region is calculated according to a coding function f ()₁＝f(x₁),h₂＝f(x₂),…，h_n＝f(x_n) Where n is the predetermined number of similar regions. The specific value of n may be determined by appropriate experiments or by comprehensive consideration of various factors such as system performance and processing rate, and will not be described herein again.

In one example, when the above-mentioned n similar regions of the current region x are acquired from the image file, the positions of the similar regions may be defined. Preferably, it may be defined that the acquired similar region cannot be located within a predetermined range centered on the current region, thereby avoiding acquiring a similar region that may have similar abnormality as the current region. On the basis of the present disclosure, those skilled in the art can appropriately determine the specific value of the predetermined range through appropriate experiments and various factors such as a priori knowledge about the possible abnormality in the image to be processed, which is not described herein again.

Subsequently, a hidden variable modification step S105 is performed in which based on the obtained respective similar areas x₁，x₂，…，x_nHidden variable h of₁,h₂,…，h_nModifying the hidden variable h of the current area x and obtaining a modified hidden variable s (h). Here, a modification function to the hidden variable is represented by s (). As an example, the modification function s () may be directly defined as a function form with parameters (such as a matrix form, in which the parameters are to be determined), or may indirectly implement a transformation relationship between the output modified hidden variable s (h) and the input hidden variable h through a multi-layer neural network or the like. As a non-limiting example, the optimization problem may be solved, for example, such that the modified hidden variable s (h) approximates the above-described hidden variable h of the similar region₁,h₂,…，h_nEtc. to determine the specific parameters of the function s (), which will be described in more detail later with reference to fig. 2 to 3.

Thereafter, a hidden variable decoding step S107 is performed, in which the modified hidden variable S (h) is decoded using a decoding function g () to obtain a reconstructed region g (S (h)) of the current region x. Since the hidden variable s (h) that has been modified based on the hidden variables of the similar regions is used, the reconstructed region g (s (h)) obtained by decoding will also be close to the image region that is similar to the current region and has a high probability of no abnormality (considering that the probability of defects existing in the entire image file is low). That is, if the current area x contains an abnormality such as a defect and is similar to it, the respective image areas x₁，x₂，…，x_nWithout the presence of anomalies, the reconstructed image region will also be substantially free of anomalies.

Next, in the abnormality determination step S109, the current region x is compared with the reconstructed region g (S (h)) obtained in step S107, and it is determined whether there is an abnormality in the current region x based on the comparison result.

As described above, the image region in which the abnormality exists is not reconstructed well (but is reconstructed to be substantially free of the image region in which the abnormality exists), and therefore when the current region in which the abnormality exists and the reconstructed region of the region are compared in the abnormality determination step S109, the difference therebetween is large, thereby enabling the image region including the defect to be detected efficiently. As an example, the difference | g(s) (h)) between the current region x and the reconstructed region g (s (h)) may be greater than a predetermined threshold Th |)₁When there is an abnormality in the current region x, it is considered. The above-mentioned threshold value Th can be appropriately determined by appropriate experiments on a priori knowledge about abnormality or the like₁The specific values of (a) are not described herein again.

As can be seen, with the information processing method of the present embodiment, regardless of whether there is an abnormality in the current region, a reconstructed region that is similar to the current region and that substantially does not include an abnormality (or a reconstructed region that is less affected by an abnormality) can be reconstructed with the self-encoder based on the similarity between the regions in the image file having the repetitive pattern, thereby making it possible to accurately detect whether there is an abnormality in the current region based on the result of comparison of such a reconstructed region with the current region. Accordingly, the information processing apparatus of the present embodiment can improve the efficiency and accuracy of abnormality detection.

Further, since a reconstructed region that is similar to the current region and that includes substantially no abnormality (or a reconstructed region that is less affected by an abnormality) can be reconstructed regardless of whether or not an abnormality exists in the current region, not only can the presence or absence of an abnormality in the current region be accurately detected, but also the position of an abnormality can be accurately detected when necessary, based on the result of comparing such a reconstructed region with the current region.

Accordingly, in a preferred embodiment, in the abnormality determining step S109, it is also possible to compare the difference of each position of the current region and the reconstructed region, and determine a position where the difference is larger than a predetermined threshold value in the comparison result as an abnormal position. Note that the specific value of the threshold may be appropriately determined by appropriate experiments and a priori knowledge about the possible abnormalities in the image to be processed, which is not described herein again. With the present preferred embodiment, in addition to the detection efficiency and accuracy of the image area where an abnormality exists in the image file, the detection efficiency and accuracy regarding the specific abnormality position in the abnormality area can be further improved.

Next, one example process of the hidden variable modification step S105 in the information processing method shown in fig. 1 will be described with reference to fig. 2.

Fig. 2 is a flowchart schematically showing one example process of a hidden variable modification step in the information processing method of fig. 1. As shown in fig. 2, in the present example, the process of the hidden variable modification step S105 may include: a step S1051 of acquiring a comprehensive similar hidden variable, which acquires a comprehensive similar hidden variable based on the acquired hidden variable of the similar region; a hidden variable difference modifying step S1053 of modifying the hidden variable of the current region based on the hidden variable difference between the hidden variable of the current region and the synthesized similar hidden variable.

With the above example processing, for example, it is possible to modify the hidden variables of the current region in such a manner that the difference (i.e., hidden variable difference) between the hidden variables of the current region and the synthesized similar hidden variables is reduced, so that the modified hidden variables approach the synthesized similar hidden variables obtained based on the hidden variables of the respective similar regions, thereby facilitating construction of a reconstructed region that is similar to the similar regions and that substantially includes no abnormalities (or is less affected by abnormalities), and further facilitating accurate detection of whether there are abnormalities in the current region.

To describe the process of fig. 2 in more detail, consider the example described above with reference to fig. 1, where the self-encoder is represented by an encoding function f () and a decoding function g (), the current region is represented by x, the hidden variable of the current region x is represented by h, and the hidden variable of the current region x is represented by x_iA similar area representing the current area x and denoted by h_iRepresenting similar regions x_iI ═ l, 2.., n, n is the number of similar regions.

With the above example, in the process of fig. 2, the integrated similar hidden variable acquisition step S1051 is first performed in which each similar area x is based on_iIs hiddenVariable h_iAnd obtaining the integrated similar hidden variables.

Preferably, the synthesized similar hidden variables may be obtained based on the hidden variables of each similar region and the similarity of each similar region to the current region. For example, it can be calculated as each similar area x according to the following formula (1)_iHidden variable h of_iThe weighted average of (a) the integrated similarity hidden variable H:

H＝w₁h₁+w₂h₂+...+w_nh_n… formula (1)

Wherein the weighting coefficient w_iCan be based on the current region x and the similar region x_iSimilarity between x and x is determined_iThe higher the similarity between, w_iThe greater the value of (A), and w_iIs normalized to satisfy the constraint condition w₁+w₂+...+w_n＝1。

Note that although the example manner of obtaining the synthesized similar hidden variables is described above in the form of equation (1), those skilled in the art will appreciate that it is by way of example only and not by way of limitation. One skilled in the art can obtain the synthetic similar hidden variables in other ways based on the present disclosure. For example, the sum of hidden variables of the similar regions may be simply divided by the total number of similar regions, i.e., obtained as the integrated similar hidden variable H', without considering the similarity between the similar regions and the current region (H ═ H)₁+h₂+...+h_n)/n。

Next, in the process of fig. 2, a hidden variable difference modifying step S1053 is performed, in which the hidden variable H of the current region x is modified based on the hidden variable difference between the hidden variable H of the current region x and the obtained synthesized similar hidden variable H, resulting in a modified hidden variable S (H).

As discussed previously, s () represents a modification function for modifying the hidden variables, and the values of the parameters in the modification function s () may be determined by solving an optimization problem based on the hidden variables of the respective similar regions, e.g., such that the obtained modified hidden variables s (h) approach the hidden variables of the similar regions.

As an example, at this time, a cost function L representing the hidden variable difference between the modified hidden variable s (H) and the synthesized similar hidden variable H as shown in equation (2) may be constructed₁So as to introduce the constraint of hidden variables of similar regions to hidden variables of current regions.

L₁Formula (2) is | (H) -H | …

By making the above-mentioned cost function L₁Minimization, the form and parameters of the modification function s () that are optimal for the hidden variable h of the current region x are determined. Accordingly, the modified hidden variable s (h) obtained based on such a modification function s () can well characterize the characteristics of each similar region, so that the reconstructed region g (s (h)) reconstructed accordingly can approach the similar region of the current region without substantially including or being substantially unaffected by a possible abnormality of the current region.

Next, another example process of the hidden variable modification step in the information processing method shown in fig. 1 will be described with reference to fig. 3.

Fig. 3 is a flowchart schematically showing one example process of a hidden variable modification step in the information processing method of fig. 1. As shown in fig. 3, the processing of the hidden variable modification step in this example differs from the processing in fig. 2 in that a reconstruction difference modification step S1055 for modifying the hidden variable of the current region based on the reconstruction difference representing the difference between the reconstructed region obtained by decoding the hidden variable of the current region with the auto-encoder and the current region is additionally included. Besides, the process of fig. 3 substantially coincides with the process of fig. 2, and therefore the description will be continued below on the basis of the description given above with reference to fig. 2, and the differences of fig. 3 from fig. 2 will be mainly described.

Here, continuing with the example described above with reference to fig. 1 and 2, the self-encoder is represented by an encoding function f () and a decoding function g (), the current region is represented by x, the hidden variable of the current region x is represented by h, and x is represented by x_iSimilar area representing the current area x, in h_iRepresenting similar regions x_iThe hidden variable(s) of (a),h denotes the integrated similarity hidden variable (i ═ l, 2.., n, n is the number of similar regions).

For the above example, after the same integrated similar hidden variable acquisition step S1051 and hidden variable difference modification step S1053 as in fig. 2, a reconstruction difference modification step S1055 is then performed in which the hidden variable of the current region is modified based on a reconstruction difference representing the difference between the reconstructed region obtained by decoding the hidden variable of the current region with an auto-encoder and the current region.

As an example, at this time, a cost function L as shown in the following formula (3) may be constructed based on the reconstruction difference₂To represent the reconstruction difference between the reconstructed region g (s (h)) obtained by decoding with the modified hidden variable s (h) and the current region x, thereby introducing the constraint of the reconstruction result on the hidden variable of the current region.

L₂Formula (3) is | g (s (h)) -x | …

Can be obtained by applying a cost function L representing the difference of hidden variables as shown in equation (2)₁Minimizing a cost function L representing a difference in reconstruction₂And minimized, thereby determining the form and parameters of the modification function s () that is optimal for the hidden variable h of the current region x. Accordingly, the modified hidden variable s (h) obtained based on such a modification function s () can well reconstruct the current region in addition to well characterizing the features of the respective similar regions, so that the reconstructed region g (s (h)) reconstructed accordingly approaches the similar regions of the current region without deviating too much from the current region (for example, if a deviation occurs in the extracted respective similar regions, such a deviation may occur).

Note that, in fig. 3, the hidden variable difference modifying step S1053 and the reconstruction difference modifying step S1055 are shown as being sequentially executed in series, but actually, the processing of the above steps may be executed in parallel, or may be executed in the same step. That is, it is possible to construct an overall loss function based on the hidden variable difference and the reconstruction difference, and to obtain the form and parameters of the optimal modification function s () in one process by optimizing the overall loss function.

In other words, preferably, the modified hidden variables may be obtained by processing the hidden variables of the current region using a neural network model optimized by minimizing a loss function constructed based on hidden variable differences and reconstruction differences as a modification function.

As an example, the above-mentioned overall loss function L constructed based on both the hidden variable difference and the reconstruction difference can be expressed as follows:

L＝L₁+λL₂(ii) (H) -H | + λ | g (s (H)) -x | … equation (4)

Wherein λ is a preset cost function L₂The coefficients of (a) may be determined appropriately according to experiments and the like, and are not described herein again.

As can be seen from the above description about the example process of fig. 3, although in the example flow of fig. 1, the hidden variable modification step S105 and the hidden variable decoding step S107 are shown to be executed in series, in the present preferred embodiment described with reference to fig. 3, the decoding result obtained by the hidden variable decoding step is actually employed in the hidden variable modification step to construct a cost function or a loss function to modify the hidden variable of the current region, and therefore, there is actually a feedback relationship of data between these two steps, as shown in fig. 4.

Fig. 4 is a flowchart schematically showing a preferred example flow of an information processing method according to an embodiment of the present disclosure, which employs an example process of the hidden variable modification step shown in fig. 3. The information processing method 400 of the present preferred embodiment differs from the information processing method 100 of fig. 1 in that, between the current region encoding step S401, the similar region encoding step S403, the hidden variable modifying step S405, the hidden variable decoding step S407, and the abnormality determining step S409, which correspond to the steps S101 to S109 in fig. 1, respectively, in addition to the relationship sequentially executed in this order similarly to that in fig. 1, there is a feedback relationship as shown by the broken-line arrow in fig. 4, that is, the decoding result is supplied as feedback data to the hidden variable modifying step S405 by the hidden variable decoding step S407 to improve the accuracy of the processing in the hidden variable modifying step S405. Otherwise, the processing of the preferred embodiment is basically the same as the processing shown in fig. 1, and thus is not described again.

The information processing method of the embodiment of the present disclosure is described above with reference to fig. 1 to 4, in which a current area of an image including a repetitive pattern is processed to determine whether there is an abnormality in the area. It is to be understood that, on the basis of the above description, the information processing method described above with reference to fig. 1 to 4 may be separately performed on other areas in the image file in addition to the processed current area to detect whether there is an abnormality in each area in the entire image file.

As an example, the abnormality detection may be performed individually for each region of the entire image, that is, the abnormality detection for each region may be performed in such a manner that the processes of the respective regions are independent of each other. In other words, the areas other than the current area in the image file may be respectively used as detection targets, and the same processing as that of the current area may be performed on each detection target to respectively determine whether there is an abnormality in each detection target.

Alternatively, in a preferred embodiment, the anomaly detection for the various regions of the entire image may be performed in a parallel manner. Such a preferred embodiment will be described below with reference to fig. 5.

Fig. 5 is a flowchart schematically showing a preferred example flow of an information processing method according to an embodiment of the present disclosure. As shown in the main flow chart on the left side in fig. 5, the information processing method 500 of the present preferred embodiment includes a current region encoding step S501, a similar region encoding step S503, a hidden variable modifying step S505, a hidden variable decoding step S507, and an abnormality determining step S509, which correspond to steps S101 to S109 in the information processing method 100 of fig. 1, respectively. The difference between the information processing method 500 of fig. 5 and the information processing method 100 of fig. 1 is mainly in the similar region encoding step S503, and thus the present preferred embodiment will be described below mainly based on the difference.

As shown in the right flowchart in fig. 5, in the similar region encoding step S503 of the present exemplary flow, the process of acquiring the similar region of the current region from the image file may include: a similar region extraction step S5030 of extracting a similar region of the current region from the image file; and steps S5031 to S5037 of performing the same processing as the current region on the extracted similar region (which may also be referred to herein as an extracted region) to obtain a reconstructed region of the extracted region and taking the reconstructed region as the acquired similar region.

That is, by executing the extracted region encoding step S5031, the similar region encoding step S5033 of the extracted region, the hidden variable modifying step S5035 of the extracted region, and the hidden variable decoding step S5037 of the extracted region, which are similar to the steps S101 to S109 in fig. 1, respectively, with the extracted region as the current region, the processing for the similar region in the main flow (as shown in steps S501 to S509 on the left side in the figure) can be executed in parallel with the main flow in the information processing method 500, as shown in the flowchart constituted by the steps S5030 to S5037 between the right side a and B in the figure.

Note that although the final abnormality determination step based on the comparison of the reconstructed region and the extracted region on the extracted region is not shown in the figure, this is omitted from the figure only for the sake of simplicity. In fact, in addition to the steps S5030 to S5037 between the right a and B in the figure, an abnormality determination step corresponding to the step S509 in fig. 1 may be performed on the extracted region, and is not described herein again.

In addition, as indicated by the leftmost and rightmost dotted arrows in the figure, the main flow (as shown in steps S501 to S509) of the information processing method 500 of fig. 5 may itself be a process executed as a "similar region" of other regions in the image file, and in the flow constituted by steps S5030 to S5037 between the right side a and B in the figure, in the similar region encoding step S5033 for the extracted region, a process similar to the flow steps S5030 to S5037 may be executed in parallel for the similar region of the extracted region, and a description thereof will not be further made.

As described above, with the processing of the present preferred embodiment, it is possible to perform abnormality detection for respective regions of the entire image in a parallel manner, thereby making it possible to improve the processing accuracy of one region with the processing result of another region.

Next, reference is made back to fig. 4. In the preferred embodiment described with reference to fig. 4, the accuracy of the processing in the hidden variable modifying processing is improved by using the decoding result obtained in the hidden variable decoding processing as feedback data. In a further preferred embodiment, the hidden variable modification and hidden variable decoding including the above feedback may be performed in an iterative manner, thereby further improving the processing accuracy of the information processing method. The process of the present preferred embodiment will be described below with reference to fig. 6.

Fig. 6 is a flowchart schematically showing a preferred example flow of an information processing method according to an embodiment of the present disclosure. As shown in fig. 6, in the information processing method 600 of the present preferred embodiment, in addition to the current region encoding step S601, the similar region encoding step S603, the hidden variable modifying step S605, the hidden variable decoding step S607, and the abnormality determining step S609, which correspond to steps S101 to S109 in the information processing method 100 of fig. 1, respectively, an iteration stop condition determining step S608 is included for determining whether or not a predetermined iteration stop condition is satisfied, and as long as the condition is not satisfied, the hidden variable modifying step S605 and the hidden variable decoding step S607 are iteratively executed with the reconstructed region obtained by the hidden variable decoding step S607 in the previous iteration as a new current region until the predetermined iteration stop condition is satisfied. Until a predetermined iteration stop condition is satisfied.

The difference between the information processing method 600 shown in fig. 6 and the information processing method 100 of fig. 1 or the information processing method 400 of fig. 4 mainly lies in the iteration stop condition determination step S608 and the relevant iteration processing, and therefore the present preferred embodiment will be described below on the basis of the description of the information processing method 400 of fig. 4 mainly based on the difference.

That is, continuing with the example described above with reference to fig. 1, 2, 4, etc., the autoencoder is represented by an encoding function f () and a decoding function g (), the current region is represented by x, the hidden variable of the current region x is represented by h, and the hidden variable of the current region x is represented by x_iSimilar area representing the current area x, in h_iRepresenting similar regions x_iWhere, i ═ l,n, n is the total number of similar regions, which may be a predetermined value, for example. In this example, a parameter T is added to indicate the number of iterations, where T is 1, 2, …, T, where T is the total number of iterations, and may be a predetermined value or a number of iterations reached when an iteration stop condition is satisfied. Accordingly, s^t() Represents the modification function, s, for modifying the hidden variable h in the t-th iteration^t(h) Representing the modified hidden variable.

With the above example, at the initial time (i.e., when t is 1), in a manner similar to the information processing method 100 described with reference to fig. 1, the current region encoding step S601, the similar region encoding step S603, the hidden variable modifying step S605, and the hidden variable decoding step S607 are sequentially performed, and the hidden variable h of the current region x, each similar region x obtained from the image file are respectively obtained by these steps_iHidden variable h of_iBased on the respective similar regions x_iHidden variable h of_iObtained modified hidden variable s¹(h) And based on the modified hidden variable s¹(h) Decoding the obtained reconstructed region g(s)¹(h))。

Next, in the iteration stop condition determination step S608, it is determined whether or not a preset iteration stop condition is satisfied. Since the initial iteration generally cannot satisfy the iteration stop condition, the hidden variable modification step S605 is returned to start the next iteration.

Next, the processing performed in the t-th iteration will be mainly described taking a case after the first iteration (t >1) as an example.

It will be appreciated that the modified hidden variable S has been obtained by the hidden variable modification step S605 in the t-1 th iteration before the t-th iteration^t-1(h) And based on the modified hidden variable S by a hidden variable decoding step S607^t-1(h) A reconstructed region g(s) is obtained^t-1(h))。

Thus, in the current t-th iteration, first, in the hidden variable modification step S605, the reconstructed region g (S) obtained in the t-1-th iteration is used^t-1(h) As newCurrent region based on each similar region x acquired before_iHidden variable h of_iModifying the hidden variable h of the current area x to obtain the modified hidden variable s in the t iteration^t(h) In that respect For example, a modification function s in the form of a neural network model may be utilized^t() Processing hidden variable h of current area x to obtain modified hidden variable s^t(h) A modification function s in the form of the neural network model^t() Is obtained by obtaining a plurality of similar regions x_iHidden variable h of_iAnd is optimized.

As an example of a specific implementation of the above-described processing, at this time, for example, processing similar to the example processing of the hidden variable modification step described with reference to fig. 2 may be performed. That is, first, a process similar to the integrated hidden variable acquisition step S1051 shown in fig. 2 may be performed, based on the respective similar regions x that have been obtained in the first process (and that do not change in the respective iterations)_iHidden variable h of_iAnd each similar region x_iWith the reconstructed region g(s) obtained in the t-1 th iteration used as the current region^t-1(h) ) of the t-th iteration, a similarity hidden variable H is obtained, for example, in the form of the integrated similarity hidden variable H of the t-th iteration^t：

H^t＝w^t ₁h₁+w^t ₂h₂+...+w^t _nh_n… equation (5)

Wherein the weighting coefficient w^t _iThe reconstructed region g(s) of the current region may be used in accordance with the current t-th iteration^t-1(h) And similar region x)_iThe similarity between them is determined, and the region g(s) is reconstructed^t-1(h) And x)_iThe higher the similarity between, w^t _iThe greater the value of (A), and w^t _iIs normalized to satisfy the constraint condition w^t ₁+w^t ₂+...+w^t _n＝1。

Next, a process similar to the hidden variable difference modification step S1053 shown in FIG. 2 may be performed, wherein the process is based onHidden variable H of current region x and synthesized similar hidden variable H obtained by current t-th iteration^tThe hidden variable difference between the previous hidden variables, the hidden variable of the current area is modified to obtain the modified hidden variable s of the t iteration^t(h)。

By way of example, at this time, a hidden variable difference cost function L similar to equation (2) may be constructed, for example^t ₁Which represents the modified hidden variable s^t(h) Analogous hidden variable H to synthesis^tTo introduce the hidden variable constraint of the similar area to the hidden variable of the current area.

L^t ₁＝‖s^t(h)-H^tII … formula (6)

By making the above-mentioned cost function L^t ₁Minimizing, determining the modification function s in the current t-th iteration^t() And accordingly obtaining the modified hidden variable s in the current t-th iteration^t(h)。

Alternatively, at this time, a reconstruction difference cost function L similar to equation (3) may also be constructed^t ₂To indicate the utilization of the modified hidden variable s^t(h) Decoding the obtained reconstructed region g(s)^t(h) ) the reconstruction difference from the current region x to introduce a constraint of the reconstruction result on the modified hidden variables of the current region.

L^t ₂＝‖g(s^t(h) Equation (7) — x | …

Preferably, based on the above equations (6) and (7), an overall loss function L similar to equation (4) can be constructed:

L^t＝L^t ₁+λL^t ₂＝‖s^t(h)-H^t‖+λ‖g(s^t(h) equation (8) — x | …

By making the above-mentioned overall loss function L^tMinimizing, determining the modification function s in the current t-th iteration^t() And accordingly obtaining the modified hidden variable s in the current t-th iteration^t(h)。

Next, at the presentContinues to a hidden variable decoding step S607 in which the modified hidden variable S obtained in the current iteration is decoded with the decoding function g () of the self-encoder^t(h) Decoding is performed to obtain a reconstructed region g(s) of the current region x^t(h))。

Then, in the current t-th iteration, the process proceeds to iteration stop condition determination step S608 to determine whether or not a predetermined iteration stop condition is satisfied. As an example, the iteration stop condition may be that the current iteration number T reaches a preset total iteration number T, or may be a modification function s determined by the current iteration^t() With the modification function s determined in the last iteration^t-1() Are substantially the same.

On the one hand, when it is determined in the iteration stop condition determination step S608 that the iteration stop condition is satisfied, the iteration process is stopped, and the process 600 proceeds to the abnormality determination step S609 to determine whether the modified hidden variable S is based on^t(h) Decoding the obtained reconstructed region g(s)^t(h) Is compared with the current area x, and it is determined whether there is an abnormality in the current area x based on the comparison result.

On the other hand, when it is judged in the iteration stop condition judgment step S608 that the iteration stop condition has not been satisfied, it is returned to the hidden variable modification step S605, the t +1 th iteration is started, and the processing of steps S605 to S608 similar to the t-th iteration is repeated until it is judged in the iteration stop condition judgment step S608 that the stop condition is satisfied.

One preferred embodiment of the information processing method of the present disclosure is described above with reference to fig. 6. In the present preferred embodiment, since the processing of hidden variable modification and hidden variable decoding is iteratively performed, the effect in modifying the hidden variable of the current region can be improved, thereby improving the accuracy of the information processing method. Next, a further preferred embodiment of the information processing method of the present disclosure will be described with reference to fig. 7.

Fig. 7 is a flowchart schematically showing a preferred example flow of an information processing method according to an embodiment of the present disclosure. As shown in fig. 7, the information processing method 700 of the present preferred embodiment includes a current region encoding step S701, a similar region encoding step S703, a hidden variable modification step S705, a hidden variable decoding step S707, an iteration stop condition determination step S708, and an abnormality determination step S709, which correspond to steps S601 to S609 in fig. 6, respectively. The only difference between the information processing method 700 in fig. 7 and the information processing method 600 in fig. 6 is that, when it is determined in the iteration stop condition determination step S708 that the determination condition is not satisfied, the three steps of the similar region encoding step S703, the hidden variable modification step S705, and the hidden variable decoding step S707 are iteratively performed to further improve the effect in modifying the hidden variable of the current region, thereby better improving the accuracy of the information processing method.

Since the difference between the information processing method 700 shown in fig. 7 and the information processing method 600 of fig. 6 is only iterative processing after initial processing (the number of iterations t >1), the description of the information processing method 700 of fig. 7 will be continued based on the example of fig. 6, and the difference of the information processing method 700 from the information processing method 600 of fig. 6 will be mainly described.

Accordingly, similarly to the example of fig. 6, it is continued to represent the self-encoder with an encoding function f () and a decoding function g (), the current region with x, the hidden variable of the current region x with h, and the number of iterations with parameter T, T being 1, 2, …, T. T is a total number of iterations, which may be a predetermined set value or may be a number of iterations reached when an iteration stop condition is satisfied. Accordingly, in this example, in x^t _iSimilar region representing the current region in the t-th iteration, in h^t _iRepresenting similar regions x in the t-th iteration^t _iAnd with s, and^t() Represents the modification function, s, for modifying the hidden variable h in the t-th iteration^t(h) A modified hidden variable is indicated, where i ═ l, 2.. and n, n is the total number of similar regions, and may be a predetermined value, for example.

For the above example, at the initial time (i.e., when t is 1), the current region coding is sequentially performed in a similar manner to the information processing method 600 described with reference to fig. 6A code step S701, a similar region encoding step S703, a hidden variable modification step S705 and a hidden variable decoding step S707, and respectively obtain a hidden variable h of a current region x and each similar region x obtained from an image file through the steps¹ _iHidden variable h of¹ _iBased on the respective similar regions x¹ _iHidden variable h of¹ _iObtained modified hidden variable s¹(h) Based on the modified hidden variable s¹(h) Decoding the obtained reconstructed region g(s)¹(h))。

Next, in the iteration stop condition determination step S708, it is determined whether or not a preset iteration stop condition is satisfied. The first iteration generally cannot satisfy the iteration stop condition, and at this time, the process 700 returns to the similar region encoding step S703 to start the next iteration.

Next, the processing performed in steps 703 to S707 of the process 700 in the t-th iteration will be described taking as an example the case after the first iteration (t > 1).

It will be appreciated that, similarly to the case of fig. 6, prior to the t-th iteration, the modified hidden variable S has been obtained in the t-1 th iteration by the hidden variable modification step S705^t-1(h) And based on the modified hidden variable S by the hidden variable decoding step S707^t-1(h) A reconstructed region g(s) is obtained^t-1(h))。

Thus, in the current t-th iteration, first, in the similar region encoding step S703, the reconstructed region g (S) obtained in the t-1-th iteration is used^t-1(h) As a new current region, the reconstructed region g(s) is obtained from the image file^t-1(h) Similar region x of (2)^t _iAnd applying a coding function f () of the self-encoder to each similar region x^t _iCoding to obtain hidden variable h of each similar area^t _i＝f(x^t _i)。

Next, in the hidden variable modification step S705, each similar region x obtained in the similar region encoding step S703 based on the current iteration^t _iHidden variables ofh^t _iModifying the hidden variable h of the current area x to obtain the modified hidden variable s in the t iteration^t(h) In that respect For example, a modification function s in the form of a neural network model may be utilized^t() Processing hidden variables of the current region to obtain modified hidden variables, a modifying function s in the form of a model of the neural network^t() Is obtained by obtaining a plurality of similar regions x^t _iHidden variable h of^t _iAnd is optimized.

As an example of a specific implementation of the above-described processing, processing similar to that in the hidden variable modification step described with reference to fig. 6 in conjunction with equations (5) to (8) may be performed, for example.

That is, first, a process similar to the integrated hidden variable acquisition step S1051 in the example process in fig. 2 may be performed, in which each similar region x obtained in the similar region encoding step S703 of the present iteration is based on^t _iHidden variable h of^t _iAnd each similar region x^t _iAnd the reconstructed region g(s) obtained in the t-1 th iteration, which is used as the current region in the current iteration^t-1(h) ) of the same order, a similar hidden variable H, for example of the form^t：

H^t＝w^t ₁h^t ₁+w^t ₂h^t ₂+...+w^t _nh^t _n… equation (5')

Wherein the weighting coefficient w^t _iThe reconstructed region g(s) of the current region may be used in accordance with the current t-th iteration^t-1(h) Similar region x obtained in the current iteration^t _iThe similarity between them is determined, and the region g(s) is reconstructed^t-1(h) And x)^t _iThe higher the similarity between, w^t _iThe greater the value of (A), and w^t _iIs normalized to satisfy the constraint condition w^t ₁+w^t ₂+...+w^t _n＝1。

Next, a process similar to the hidden variable difference modifying step S1053 in the example process of fig. 2 may be performed, in which the hidden variable H based on the current region x and the synthesized similar hidden variable H obtained at the current t-th iteration^tThe hidden variable difference between the previous hidden variables, the hidden variable of the current area is modified to obtain the modified hidden variable s of the t iteration^t(h)。

As an example, at this time, for example, a hidden variable difference cost function L as shown in equation (6) may be constructed^t ₁＝‖s^t(h)-H^tII which represents the modified hidden variable s^t(h) Analogous hidden variable H to synthesis^tTo introduce the hidden variable constraint of the similar area to the hidden variable of the current area.

By making the above-mentioned hidden variable difference cost function L^t ₁Minimizing, determining the modification function s in the current t-th iteration^t() And accordingly obtaining the modified hidden variable s in the current t-th iteration^t(h)。

Alternatively, at this time, a reconstruction difference cost function L as shown in equation (7) may also be constructed^t ₂＝‖g(s^t(h) X |) to represent the utilization of the modified hidden variable s^t(h) Decoding the obtained reconstructed region g(s)^t(h) ) the reconstruction difference from the current region x to introduce a constraint of the reconstruction result on the modified hidden variables of the current region.

Preferably based on the cost function L described above^t ₁And L^t ₂An overall loss function L as shown in equation (8) can be constructed^t＝L^t ₁+λL^t ₂＝‖s^t(h)-H^t‖+λ‖g(s^t(h))-x‖。

Next, at the presentContinues to a hidden variable decoding step S707 in which the modified hidden variable S obtained in the current iteration is decoded using the decoding function g () of the self-encoder^t(h) Decoding is performed to obtain a reconstructed region g(s) of the current region x^t(h))。

Then, in the current t-th iteration, the process proceeds to iteration stop condition determination step S708 to determine whether or not a predetermined iteration stop condition is satisfied. As an example, the iteration stop condition may be that the current iteration number T reaches a preset total iteration number T, or may be a modification function s determined by the current iteration^t() With the modification function s determined in the last iteration^t-1() Are substantially the same.

On the one hand, when it is judged in the iteration stop condition judgment step S708 that the iteration stop condition is satisfied, the iteration process is stopped, and the process 700 proceeds to the abnormality judgment step S709 where the modified hidden variable S is modified based on^t(h) Decoding the obtained reconstructed region g(s)^t(h) Is compared with the current area x, and it is determined whether there is an abnormality in the current area x based on the comparison result.

On the other hand, when it is judged in the iteration stop condition judgment step S708 that the iteration stop condition has not been satisfied, the process returns to the similar region modification step S703, and the t +1 th iteration is started to repeat the processes of steps S703 to S708 similar to the t-th iteration until it is judged in the iteration stop condition judgment step S708 that the iteration stop condition is satisfied.

One preferred embodiment of the information processing method of the present disclosure is described above with reference to fig. 7. In the present preferred embodiment, since the processes of the similar region encoding, hidden variable modification, and hidden variable decoding are iteratively performed, the effect in modifying the hidden variable of the current region can be improved, thereby improving the accuracy of the information processing method.

Note that the preferred embodiment of performing the respective iterative processes for the current region described above with reference to fig. 6 and 7 may be combined with the embodiment of performing parallel processing on respective regions in the entire image file described hereinbefore with reference to fig. 5.

More specifically, in the example flow of fig. 5, the main flow 500 on the left side may additionally include an iteration stop condition decision step between the hidden variable decoding step S507 and the abnormality decision step S509, and the processing of the hidden variable modification step S505 and the hidden variable decoding step S507 may be iteratively performed in a similar manner to the flow 600 in fig. 6 when this step decides that the iteration stop condition has not been satisfied, or the processing of the similar region encoding step S503, the hidden variable modification step S505, and the hidden variable decoding step S507 may be iteratively performed in a similar manner to the flow 700 in fig. 7 when this step decides that the iteration stop condition has not been satisfied. Similarly, in the right flow of fig. 5 for the extracted similar region (extracted region), an additional iteration stop condition determining step may be included after the hidden variable decoding step S5037 of the extracted region, and the step returns to the similar region encoding step S5033 of the extracted region or the hidden variable modifying step S5035 of the extracted region according to the determination result of the step to perform the iterative process similar to that in fig. 6 or fig. 7, which is not described again here.

The specific embodiments of the information processing method according to the present disclosure and advantageous effects thereof are described above with reference to fig. 1 to 7. In addition, according to the present disclosure, a corresponding information processing apparatus is also provided. These means will be described below with reference to fig. 8 to 11.

Fig. 8 is a schematic block diagram schematically showing one example structure of an information processing apparatus according to an embodiment of the present disclosure. As shown in fig. 8, the information processing apparatus 800 may include: a current region encoding unit 801, configured to encode a current region of an image file serving as a detection target by using a pre-trained auto-encoder to obtain a hidden variable of the current region; a similar region encoding unit 802, configured to acquire a similar region of a current region from an image file, and encode each similar region by using an auto-encoder to acquire a hidden variable of each similar region; a hidden variable modifying unit 803, configured to modify a hidden variable of the current region based on the obtained hidden variable of the similar region; a hidden variable decoding unit 804, which decodes the modified hidden variable by using the self-encoder to obtain a reconstructed region of the current region; and an abnormality determination unit 805 for comparing the current region with the reconstructed region and determining whether there is an abnormality in the current region based on the comparison result.

The above-described information processing apparatus 800 and its respective units shown in fig. 8 may perform, for example, the operations and/or processes of the information processing method of the present disclosure and its respective steps described above with reference to fig. 1 and achieve similar effects, and a repeated description thereof will not be provided herein.

Fig. 9 is a schematic block diagram schematically showing one example structure of a hidden variable modification unit 803 in the information processing apparatus of fig. 8. As shown in fig. 9, in the present example, the hidden variable modification unit 803 may include: a comprehensive similar hidden variable obtaining unit 8031, configured to obtain a comprehensive similar hidden variable based on the obtained hidden variable of the similar region; a hidden variable difference modifying unit 8032, configured to modify the hidden variable of the current region based on a hidden variable difference between the hidden variable of the current region and the similar hidden variable.

The hidden variable modification unit 803 and each unit thereof in fig. 9 may perform, for example, the operations and/or processes of the hidden variable modification step S105 and each step thereof in the information processing method of the present disclosure described above with reference to fig. 2 and achieve similar effects, and a repeated description thereof will not be provided.

Fig. 10 is a schematic block diagram schematically showing another exemplary structure of the hidden variable modification unit 803 in the information processing apparatus of fig. 8. As shown in fig. 10, in the present example, the hidden variable modification unit 803 additionally includes, in addition to the

same units

8031, 8032 as in fig. 9: a reconstruction difference modifying unit 8033 for modifying the hidden variable of the current region based on a reconstruction difference representing a difference between the reconstructed region obtained by decoding the hidden variable of the current region with the auto-encoder and the current region.

The hidden variable modification unit 803 and each unit thereof in fig. 10 may perform, for example, the operations and/or processes of the hidden variable modification step S105 and each step thereof in the information processing method of the present disclosure described above with reference to fig. 3 and achieve similar effects, and a repeated description thereof will not be provided.

In addition, it is understood that, with the information processing apparatus 800 and its respective units described above with reference to fig. 8 to 10, the processes in the preferred embodiment or preferred example flow of the information processing method described above with reference to fig. 4 to 5 can be performed, and a repeated description thereof will not be made here.

Fig. 11 is a schematic block diagram schematically showing one preferred example structure of an information processing apparatus according to an embodiment of the present disclosure. As shown in fig. 11, in the information processing apparatus 1100 of the preferred embodiment, in addition to the current region encoding unit 1101, the similar region encoding unit 1102, the hidden variable modification unit 1103, the hidden variable decoding unit 1104, and the abnormality determination unit 1105 which correspond to the

units

801 and 805 in fig. 8, respectively, an iteration stop condition determination unit 1106 is included.

In the present preferred embodiment, the iteration stop condition decision unit 1106 is configured to determine whether a predetermined iteration stop condition is satisfied. As long as this condition is not satisfied, the information processing apparatus 1100 may take the reconstructed region obtained by the hidden variable decoding unit 1107 in the previous iteration as a new current region in a similar manner as in the information processing method 600 described above with reference to fig. 6, so that the hidden variable modification unit 1105 and the hidden variable decoding unit 1107 iteratively perform their respective processes until a predetermined iteration stop condition is satisfied.

Alternatively, as long as the iteration stop condition determination unit 1106 determines that the predetermined iteration stop condition has not been satisfied, the information processing apparatus 1100 may take the reconstructed region obtained by the hidden variable decoding unit 1107 in the previous iteration as a new current region in a similar manner to that in the information processing method 700 described above with reference to fig. 7, so that the similar region encoding unit 1102, the hidden variable modification unit 1105, and the hidden variable decoding unit 1107 iteratively perform their respective processes until the predetermined iteration stop condition is satisfied.

Accordingly, the information processing apparatus 1100 of the present preferred embodiment and its respective units can, for example, perform the operations and/or processes of the respective steps of the preferred embodiment of the information processing method of the present disclosure described above with reference to fig. 6 or 7 and achieve similar effects, and a repetitive description thereof will not be made herein.

The above describes various embodiments of an information processing method and apparatus that the present disclosure can provide, and advantageous effects thereof. One possible hardware configuration that may be used to implement the methods and apparatus will be described below with reference to fig. 12.

In fig. 12, a Central Processing Unit (CPU)1201 executes various processes in accordance with a program stored in a Read Only Memory (ROM)1202 or a program loaded from a storage section 1208 to a Random Access Memory (RAM) 1203. In the RAM 1203, data necessary when the CPU 1201 executes various processes and the like is also stored as necessary. The CPU 1201, the ROM 1202, and the RAM 1203 are connected to each other via a bus 1204. An input/output interface 1205 is also connected to bus 1204.

The following components are also connected to the input/output interface 1205: an input section 1206 (including a keyboard, a mouse, and the like), an output section 1207 (including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker, and the like), a storage section 1208 (including a hard disk, and the like), and a communication section 1209 (including a network interface card such as a LAN card, a modem, and the like). The communication section 1209 performs communication processing via a network such as the internet. A driver 1210 may also be connected to the input/output interface 1205 as desired. A removable medium 1211 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like may be mounted on the drive 1210 as necessary, so that a computer program read out therefrom can be installed into the storage section 1208 as necessary.

In addition, the present disclosure also provides a program product storing machine-readable instruction codes. When the instruction codes are read and executed by a machine, the information processing method according to the embodiment of the disclosure can be executed. Accordingly, various storage media such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. for carrying such a program product are also included in the disclosure of the present disclosure.

That is, the present disclosure also proposes a storage medium storing machine-readable instruction codes, which, when read and executed by a machine, can cause the machine to execute the above-described information processing method according to an embodiment of the present disclosure.

The storage medium may include, for example, but is not limited to, a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like.

In the foregoing description of specific embodiments of the disclosure, features described and/or illustrated with respect to one embodiment may be used in the same or similar manner in one or more other embodiments, in combination with or instead of the features of the other embodiments.

Furthermore, the methods of the embodiments of the present disclosure are not limited to being performed in the chronological order described in the specification or shown in the drawings, and may be performed in other chronological orders, in parallel, or independently. Therefore, the order of execution of the methods described in this specification does not limit the technical scope of the present disclosure.

Further, it is apparent that the respective operational procedures of the above-described method according to the present disclosure can also be implemented in the form of computer-executable programs stored in various machine-readable storage media.

Moreover, the object of the present disclosure can also be achieved by: a storage medium storing the above executable program code is directly or indirectly supplied to a system or an apparatus, and a computer or a Central Processing Unit (CPU) in the system or the apparatus reads out and executes the program code.

At this time, as long as the system or the apparatus has a function of executing a program, the embodiments of the present disclosure are not limited to the program, and the program may also be in any form, for example, an object program, a program executed by an interpreter, a script program provided to an operating system, or the like.

Such machine-readable storage media include, but are not limited to: various memories and storage units, semiconductor devices, magnetic disk units such as optical, magnetic, and magneto-optical disks, and other media suitable for storing information, etc.

In addition, the client information processing terminal can also implement the embodiments of the present disclosure by connecting to a corresponding website on the internet, and downloading and installing computer program codes according to the present disclosure into the information processing terminal and then executing the program.

In summary, according to the embodiments of the present disclosure, the present disclosure provides the following schemes, but is not limited thereto:

an information processing apparatus capable of being used for detecting an abnormality in an image file containing a repetitive pattern, the apparatus comprising:

a processor configured to:

encoding a current region of an image file serving as a detection target by using a pre-trained self-encoder to obtain a hidden variable of the current region;

acquiring similar areas of a current area from an image file, and encoding each similar area by using an autoencoder to acquire hidden variables of each similar area;

modifying the hidden variable of the current area based on the acquired hidden variable of the similar area;

decoding the modified hidden variable by using an auto-encoder to obtain a reconstructed region of the current region; and

and comparing the current area with the reconstruction area, and judging whether the current area has abnormality or not based on the comparison result.

Scheme 2. the information processing apparatus of scheme 1, wherein the processor is further configured to modify the hidden variable of the current region by:

acquiring a comprehensive similar hidden variable based on the acquired hidden variable of the similar area;

and modifying the hidden variables of the current area based on the hidden variable difference between the hidden variables of the current area and the similar hidden variables.

Scheme 3. the information processing apparatus of scheme 2, wherein the processor is further configured to: and obtaining a comprehensive similar hidden variable based on the hidden variable of each similar area and the similarity between each similar area and the current area.

Scheme 4. the information processing apparatus of any of schemes 1 to 3, wherein the processor is further configured to: modifying the hidden variable of the current region and decoding the modified hidden variable based on the obtained hidden variable of the similar region are iteratively performed with the obtained reconstructed region as a new current region until a predetermined iteration stop condition is satisfied.

Scheme 5. the information processing apparatus of any of schemes 1 to 3, wherein the processor is further configured to: with the obtained reconstructed region as a new current region, iteratively performing the steps of obtaining the similar region and its hidden variables, modifying the hidden variables of the current region based on the obtained hidden variables of the similar region, and decoding the modified hidden variables until a predetermined iteration stop condition is satisfied.

Scheme 6. the information processing apparatus of scheme 2 or 3, wherein the processor is further configured to: the hidden variable of the current region is also modified based on a reconstruction difference representing a difference between the current region and a reconstructed region obtained by decoding the hidden variable of the current region using an auto-encoder.

Scheme 7. the information processing apparatus of scheme 6, wherein the processor is further configured to process the hidden variables of the current region using a neural network model optimized by minimizing a loss function constructed based on hidden variable differences and reconstructed differences to obtain modified hidden variables.

Scheme 8 the information processing apparatus according to any one of schemes 1 to 3, wherein the processor is further configured to obtain the similar area of the current area by:

extracting a similar area of the current area from the image file; and

the same processing as that of the current region is performed on the extracted similar region to obtain a reconstructed region of the similar region, and the reconstructed region is taken as the acquired similar region.

Scheme 9. the information processing apparatus of any of schemes 1 to 3, wherein the processor is further configured to: the difference of each position of the current region and the reconstructed region is compared, and a position where the difference is larger than a predetermined threshold in the comparison result is determined as an abnormal position.

An information processing method for detecting an abnormality in an image file containing a repetitive pattern, the method comprising:

Scheme 11. the information processing method according to scheme 10, wherein modifying the hidden variable of the current region based on the obtained hidden variable of the similar region comprises:

Scheme 12. the information processing method according to scheme 11, wherein the integrated similar hidden variables are obtained based on the hidden variables of each similar region and the similarity between each similar region and the current region.

Scheme 13. the information processing method according to any one of schemes 10 to 12, further comprising:

modifying the hidden variable of the current region and decoding the modified hidden variable based on the obtained hidden variable of the similar region are iteratively performed with the obtained reconstructed region as a new current region until a predetermined iteration stop condition is satisfied.

Scheme 14. the information processing method according to any one of schemes 10 to 12, further comprising:

with the obtained reconstructed region as a new current region, iteratively performing the steps of obtaining the similar region and its hidden variables, modifying the hidden variables of the current region based on the obtained hidden variables of the similar region, and decoding the modified hidden variables until a predetermined iteration stop condition is satisfied.

Scheme 15. the information processing method according to scheme 11 or 12, wherein the hidden variable of the current region is further modified based on a reconstruction difference representing a difference between a reconstructed region obtained by decoding the hidden variable of the current region with an auto-encoder and the current region.

Scheme 16. the information processing method of scheme 15, wherein the modified hidden variables are obtained by processing the hidden variables of the current region using a neural network model optimized by minimizing a loss function constructed based on hidden variable differences and reconstructed differences.

Scheme 17 the information processing method according to any one of schemes 10 to 12, wherein the acquiring of the similar region of the current region from the image file includes:

extracting a similar area of the current area from the image file; and

The information processing method according to any one of claims 10 to 12, wherein the step of determining whether there is an abnormality includes:

the difference of each position of the current region and the reconstructed region is compared, and a position where the difference is larger than a predetermined threshold in the comparison result is determined as an abnormal position.

Scheme 19. the information processing method according to any one of schemes 10 to 12, further comprising:

and respectively taking other areas except the current area in the image file as detection targets, and performing the same processing as the current area on each detection target to judge whether each detection target has abnormality.

Scheme 20. the information processing method according to any one of schemes 10 to 12, wherein the repetitive pattern includes a pattern that repeats periodically.

Finally, it is also noted that, in the present disclosure, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements may include not only those elements but other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

While the disclosure has been disclosed by the description of specific embodiments thereof, it will be appreciated that those skilled in the art will be able to devise various modifications, improvements, or equivalents of the disclosure within the spirit and scope of the appended claims. Such modifications, improvements and equivalents are intended to be included within the scope of the present disclosure as claimed.

Claims

1. An information processing apparatus that can be used to detect an abnormality in an image file containing a repetitive pattern, the apparatus comprising:

a processor configured to:

2. The information processing apparatus of claim 1, wherein the processor is further configured to modify the hidden variable of the current region by:

3. The information processing apparatus of claim 2, wherein the processor is further configured to: and obtaining a comprehensive similar hidden variable based on the hidden variable of each similar area and the similarity between each similar area and the current area.

4. The information processing apparatus of any one of claims 1 to 3, wherein the processor is further configured to: modifying the hidden variable of the current region and decoding the modified hidden variable based on the obtained hidden variable of the similar region are iteratively performed with the obtained reconstructed region as a new current region until a predetermined iteration stop condition is satisfied.

5. The information processing apparatus of any one of claims 1 to 3, wherein the processor is further configured to: with the obtained reconstructed region as a new current region, iteratively performing the steps of obtaining the similar region and its hidden variables, modifying the hidden variables of the current region based on the obtained hidden variables of the similar region, and decoding the modified hidden variables until a predetermined iteration stop condition is satisfied.

6. The information processing apparatus of claim 2 or 3, wherein the processor is further configured to: the hidden variable of the current region is also modified based on a reconstruction difference representing a difference between the current region and a reconstructed region obtained by decoding the hidden variable of the current region using an auto-encoder.

7. The information processing apparatus of claim 6, wherein the processor is further configured to process the hidden variables of the current region using a neural network model optimized by minimizing a loss function constructed based on hidden variable differences and reconstructed differences to obtain the modified hidden variables.

8. The information processing apparatus according to any one of claims 1 to 3, wherein the processor is further configured to acquire the similar area of the current area by:

extracting a similar area of the current area from the image file; and

9. The information processing apparatus of any one of claims 1 to 3, wherein the processor is further configured to: the difference of each position of the current region and the reconstructed region is compared, and a position where the difference is larger than a predetermined threshold in the comparison result is determined as an abnormal position.

10. An information processing method for detecting an anomaly in an image file containing a repetitive pattern, the method comprising: