Intelligent information auditing method and intelligent auditing system

Abstract

The invention relates to an intelligent information auditing method and an intelligent auditing system, and the method comprises the steps: S1, obtaining an information image set, classifying all information images in the information image set, and obtaining a classification result; s2, according to the classification result, corresponding to a more matched information auditing process; and S3, completing an information auditing process and obtaining an information auditing result. In information image auditing work in the field of intelligent risk control and intelligent credit auditing, the currently common OCR technology is mainly based on a detection and recognition model, and still has defects from an actual auditing result; based on this, according to the scheme provided by the invention, after the classification result is obtained, a more matched information auditing process corresponds, and the accuracy degree of information auditing can be remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent risk control and intelligent credit investigation, and in particular to an intelligent information auditing method and an intelligent auditing system. BACKGROUND

[0002] In the information image auditing work in the field of intelligent risk control and intelligent credit investigation, the currently commonly used OCR technology is mainly based on detection and recognition models. From the actual auditing results, there are still deficiencies. Specifically, the detection model is difficult to accurately locate the boundary of blurred or occluded text, resulting in incomplete text area obtained by the recognition model; the recognition model processes special fonts and handwritten text slowly, and frequently needs manual review, which slows down the credit investigation process.

[0003] Therefore, it is necessary to design an intelligent information auditing method and an intelligent auditing system to solve the above technical problems. SUMMARY

[0004] The present application provides an intelligent information auditing method and an intelligent auditing system. After obtaining the classification result, a more matched information auditing process is corresponded, which can significantly improve the accuracy of information auditing.

[0005] In a first aspect, the present application provides an intelligent information auditing method, which comprises: S1, obtaining an information image set, classifying all information image pieces in the information image set, and obtaining a classification result; S2, according to the classification result, a more matched information auditing process is corresponded; S3, completing the information auditing process and obtaining an information auditing result.

[0006] Further technical solutions are that the step S1 comprises: S11, obtaining an information image set, classifying all information image pieces in the information image set according to the recognizable degree, obtaining a conventional image set with a recognizable degree higher than a preset threshold, and obtaining a special image set other than the conventional image set.

[0007] Further technical solutions are that the step S2 comprises: S21, according to the classification result, the conventional image set is corresponded to a first conventional information auditing process; S22, according to the classification result, the special image set is corresponded to a second special information auditing process with a higher operation complexity than the first conventional information auditing process.

[0008] A further technical solution is that the first regular information review process comprises: converting the regular image set into a first text detection feature map based on the cropped full convolutional network text detection model; converting the first text detection feature map into a binary image of a text region by using a binary module, and then obtaining position information of a text box by using a post-processing algorithm; using a quantized deep separable convolution as a feature extraction network to extract a second text recognition feature map, inputting the second text recognition feature map into a CTC layer for recognition, finding an optimal alignment path between an input feature sequence and a target text sequence, and finally obtaining a first predicted text sequence based on the optimal alignment path to complete the review process.

[0009] A further technical solution is that the second special information review process comprises: performing bottom layer visual feature extraction on a specific image set, and performing feature aggregation and dimension compression to obtain a first vector sequence; setting a uniform target dimension as a preset value, performing dimension conversion on the first vector sequence by using a linear embedding layer to obtain a second vector sequence; the image features of the second vector sequence are consistent with the text feature dimension; adding position encoding to the second vector sequence by using a sine-cosine position encoding method to obtain a first predicted text sequence, and then completing the review process.

[0010] A further technical solution is that the step S3 comprises: S61, obtaining N specific image pieces with a recognition posteriori probability not greater than a preset threshold in the second special information review process as N image pieces to be processed; S62, calling a past first image piece recognition result to determine whether there is an image piece with a similarity greater than the posteriori probability in the past first image piece recognition result; and S63, if there is, calling a recognition result corresponding to the image piece with the similarity greater than the posteriori probability as a first part of the information review result.

[0011] A further technical solution is that after the step S63, the method further comprises: if there is not, obtaining a trusted recognition result input by a human being, establishing a corresponding relationship between the image piece to be processed and the trusted recognition result, adding the corresponding relationship to the past first image piece recognition result to form a second part of the information review result, and combining the first part of the information review result and the second part of the information review result to form the information review result.

[0012] In the existing information image auditing work, there are many deficiencies, for example, there are regular images that are easy to identify and specific images that are not easy to identify (such as blurred printing and non-standard handwriting) in a picture. At this time, the identification result is not ideal by using the conventional intelligent auditing method, especially the identification clarity cannot be guaranteed. However, in the case of limited computing resources, processing such pictures containing "specific images that are not easy to identify" alone will slow down the auditing efficiency and increase the auditing cost. Therefore, the present application proposes an intelligent information auditing method, which first obtains two classification results, and then uses conventional intelligent auditing for the regular image set and uses special intelligent auditing with higher computational complexity for the specific image set. In the case of specific images that are not easy to identify, the intelligent picture auditing is completed at one time, which significantly improves the auditing efficiency and significantly reduces the auditing cost compared with the prior art.

[0013] In summary, in the information image auditing work in the field of intelligent risk control and intelligent credit investigation, the current commonly used OCR technology is mainly based on detection and recognition models. From the actual auditing results, there are still deficiencies: the detection model cannot accurately locate the boundary of blurred or occluded text, resulting in incomplete text area obtained by the recognition model; the recognition model processes special fonts and handwritten text slowly, and frequent manual review slows down the credit investigation process; based on this, the present application proposes an intelligent information auditing method and an intelligent auditing system. After obtaining the classification results, the more matched information auditing process can significantly improve the accuracy of information auditing. BRIEF DESCRIPTION OF DRAWINGS

[0014] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0016] Figure 1 A flowchart of the intelligent information auditing method provided by the embodiment of the present application.

[0017] Figure 2 Another flowchart of the intelligent information auditing method provided by the embodiment of the present application.

[0018] Figure 3 A framework diagram of the electronic device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0019] With reference to the drawings and embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are some of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of the present application.

[0020] It should be understood that the terms "comprise" and "include" as used in the specification and the appended claims indicate the presence of the described features, integers, steps, operations, elements, and / or components but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0021] It should also be understood that the terms used in the present application specification are only for the purpose of describing particular embodiments and are not intended to limit the present application. As used in the present application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should be further understood that the term "and / or" as used in the present application specification and the appended claims means one or any combination of the associated listed items and all possible combinations thereof, and includes these combinations.

[0023] As used in the present application specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting" depending on the context. Similarly, the phrase "if it is determined" or "if [a described condition or event] is detected" can be interpreted as meaning "upon determining" or "in response to determining" or "upon detecting [the described condition or event]" or "in response to detecting [the described condition or event]" depending on the context.

[0024] In the present application specification and the appended claims, there can be multiple expressions for the same technical features or professional terms, such as different expressions in the form of superordinate generalization, subordinate limitation or synonymous replacement; those skilled in the art can understand the technical meanings of the same substance indicated by different expressions based on their professional knowledge and in combination with the whole content of the specification and the drawings; the differences between different expressions only reflect the diversity in the literal level, and do not constitute substantial modification or limitation of the technical solutions, and will not affect the certainty of the patent claim protection scope and the full disclosure of the technical content in the specification.

[0025] Embodiment 1 Please refer to Figures 1 to 2 As shown in the figure, an intelligent information auditing method is provided in the embodiments of the present application. For details, please refer toFigure 1 As shown, an intelligent information auditing method includes: S1, obtaining an information image set, classifying all information image pieces in the information image set, and obtaining a classification result; S2, according to the classification result, corresponding to a more matched information auditing process; S3, completing the information auditing process and obtaining an information auditing result. In the information image piece auditing work in the field of intelligent risk control and intelligent credit investigation, the currently commonly used OCR technology is mainly based on detection and recognition models, and there are still deficiencies from the actual auditing results; based on this, the scheme described in the application can significantly improve the accuracy of information auditing after corresponding to a more matched information auditing process after obtaining the classification result.

[0026] A further technical solution is that the step S1 includes: S11, obtaining an information image set, classifying all information image pieces in the information image set according to the recognizable degree, obtaining a conventional image set with a recognizable degree higher than a preset threshold, and obtaining a specific image set other than the conventional image set.

[0027] A further technical solution is that the step S2 includes: S21, according to the classification result, corresponding to a first conventional information auditing process matched with the conventional image set; S22, according to the classification result, corresponding to a second special information auditing process with a higher operation complexity than the first conventional information auditing process for the specific image set.

[0028] A further technical solution is that the first conventional information auditing process includes: based on a cropped full convolution network text detection model, converting the conventional image set into a first text detection feature map; using a binarization module to convert the first text detection feature map into a binary image of a text region, and then obtaining position information of a text box through a post-processing algorithm; using a quantized deep separable convolution as a feature extraction network and extracting a second text recognition feature map, inputting the second text recognition feature map into a CTC layer for recognition, finding an optimal alignment path between an input feature sequence and a target text sequence, and finally obtaining a first predicted text sequence based on the optimal alignment path to complete the auditing process. In the above scheme, the position information of the text box is a key bridge connecting "text detection" and "text recognition", and directly determines the accuracy and efficiency of the first conventional information auditing process (for conventional images); accurate framing from conventional images obtains an effective text region, and excludes irrelevant interference such as background and noise; for example, in the auditing of conventional images such as identity cards and contracts, the text box of the key field such as the name and the certificate number can be directly positioned, avoiding misrecognition of non-text region content by the recognition model.

[0029] In the above scheme, the post-processing algorithm is deeply optimized based on the post-processing process of differentiable binarization, a graph neural network is introduced, and after generating the basic threshold map, a text region relationship graph is constructed in the spatial dimension; through message propagation between nodes, each text instance can learn and share the local optimal threshold of its neighborhood, realizing the adaptive allocation of threshold in the semantically similar area; after applying the post-processing algorithm, the robustness of the model to uneven illumination, low contrast and complex background is significantly improved, and a purer and more complete binarized image is generated from the source.

[0030] Further, the first text detection feature map solves the problem of fuzzy text boundary positioning in traditional detection through "full convolutional network + spatial position feature", ensuring accurate text box position information; the second text recognition feature map avoids the problem of excessive calculation and slow recognition caused by "indiscriminate processing of the entire image" in traditional recognition through "text box local input + character content feature", and focuses on text details to improve recognition accuracy, forming a "detection + recognition closed loop" to adapt to the "efficiency + accuracy" needs of the risk control credit review scene. Further, the first predicted text sequence directly provides the content of the risk control credit review, enabling subsequent information review to be based on clear text data, so its generation means the achievement of the common goal of "detection + recognition", and the two links are naturally completed; wherein the predicted text sequence is the image text information finally obtained through the text detection + recognition algorithm model.

[0031] Further, the second special information review process includes: extracting bottom-level visual features from a specific image set, and aggregating and dimensionally compressing the features to obtain a first vector sequence; setting a uniform target dimension to a preset value, and converting the first vector sequence to a second vector sequence through a linear embedding layer; the image features and text features of the second vector sequence have consistent dimensions; adding position encoding to the second vector sequence using a sine-cosine position encoding method to obtain a first predicted text sequence, thereby completing the review process. In the above scheme, the first vector sequence and the second vector sequence are both associated with "embedding vectors", and the vector sequence is an ordered set composed of multiple embedding vectors in a specific logical order; further, the embedding vector is a single-dimensional fixed numerical vector converted from image local feature raw heterogeneous data through a Word2Vec embedding model, and its core function is to map unstructured information such as pixels and characters into structured and computable low / high-dimensional numerical representations, and the vector dimension needs to meet the unified 512-dimensional multi-modal semantic alignment.

[0032] Further, the vector sequence is an ordered set composed of a plurality of embedding vectors in a specific logical order, and the core function is to retain the structural association of the original data (such as image spatial position, text semantic order), providing "context-dependent information" for the subsequent MobileViT model; the sequence length is determined by the structural characteristics of the original data, and each element is an embedding vector with consistent dimensions, which is to ensure that the subsequent recognized text information is one-to-one corresponding to the image piece. Further, the image modality obtains image modality embedding vectors with image spatial position information by using the sine-cosine position encoding method; this part is the processing of the image modality, plus the splicing of the text modality and the multi-modal sequence, and finally the "second predicted text sequence" is obtained, that is, the predicted text information on the image piece.

[0033] In the existing information image piece review work, there are many deficiencies, for example, there are easy-to-identify regular images and difficult-to-identify specific images (such as printed blur and non-standard handwriting) in a picture, at this time, using the conventional intelligent review method, the recognition result is not ideal, especially the recognition clarity cannot be guaranteed. However, in the case of limited computing resources, processing such pictures containing "difficult-to-identify specific images" alone can easily slow down the review efficiency and increase the review cost. Therefore, the present application proposes an intelligent information review method, which first obtains two classification results, and then uses conventional intelligent review for the regular image set and special intelligent review with higher computational complexity for the specific image set, which realizes one-time completion of picture intelligent review in the face of difficult-to-identify specific images. Compared with the prior art, on the one hand, the review efficiency is significantly improved, and on the other hand, the review cost is significantly reduced.

[0034] A further technical solution is that the step S3 comprises: S61, obtaining N specific image pieces with a recognition confidence not greater than a preset threshold in the second special information review process as N image pieces to be processed; S62, calling the past first image piece recognition result to determine whether there is an image piece with a similarity greater than the confidence in the past first image piece recognition result; S63, if there is, calling the recognition result corresponding to the image piece with a similarity greater than the confidence as the first part of the information review result.

[0035] A further technical solution is that after the step S63, it further comprises: if there is not, obtaining a trusted recognition result input by a human being, establishing a correspondence between the image piece to be processed and the trusted recognition result, adding the correspondence to the past first image piece recognition result to form a second part of the information review result, and combining the first part of the information review result to form the information review result.

[0036] In a second aspect, the application provides an intelligent auditing system, which is used to implement the intelligent information auditing method according to any of the above embodiments.

[0037] In summary, in the information image auditing work in the field of intelligent risk control and intelligent credit investigation, the currently commonly used OCR technology is mainly based on detection and recognition models, and there are still deficiencies in the actual auditing results: the detection model is difficult to accurately position the boundary of blurred or occluded text, resulting in incomplete text area obtained by the recognition model; the recognition model processes special fonts and handwritten text slowly, and frequent manual review slows down the credit investigation process; therefore, the application provides an intelligent information auditing method and an intelligent auditing system, which correspond to a more matched information auditing process after obtaining the classification result, and can significantly improve the accuracy of information auditing.

[0038] Embodiment 2 Please refer to Figure 3 , Figure 3 A block diagram of an electronic device is provided. The electronic device can be a terminal or a server, wherein the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a personal digital assistant, a wearable device, and an electronic device with a communication function. The electronic device includes a processor 111, a communication interface 112, a memory 113, and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114.

[0039] The memory 113 is used to store a computer program.

[0040] In an embodiment of the application, the processor 111 is used to execute the program stored in the memory 113, and implement the method provided in any of the above method embodiments.

[0041] It should be understood that, in the embodiments of the application, the processor 111 can be a central processing unit (CPU), and the processor 111 can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0042] Those skilled in the art can understand that all or part of the processes in the method for implementing the above embodiments can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a storage medium, which is a computer readable storage medium. The computer program is executed by at least one processor in the computer system to implement the process steps of the above-mentioned embodiment of the method.

[0043] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in a general manner. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions, but such implementation should not be considered beyond the scope of the present application.

[0044] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and actual implementation can have another division manner. For example, a unit or component can be combined or integrated into another system, or some features can be ignored or not executed.

[0045] The steps in the method of the embodiments of the present application can be adjusted, combined and deleted in sequence according to actual needs. The units in the device of the embodiments of the present application can be combined, divided and deleted according to actual needs. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0046] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a storage medium. Based on such understanding, the technical solutions of the present application essentially or say the parts that make contributions to the prior art, or all or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for making a computer device (which can be a personal computer, a terminal or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application.

[0047] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0048] Obviously, various modifications and changes can be made to the present application without departing from the spirit and scope thereof. Accordingly, it is intended that all such modifications and changes be included within the scope of the application as defined by the following claims and their equivalents.

[0049] The above description is merely that of the preferred embodiments of the application, but the protection scope of the application is not limited thereto. Any modifications or replacements within the technical scope disclosed by the application can be easily thought of by those skilled in the art, and these modifications or replacements shall be included in the protection scope of the application. Therefore, the protection scope of the application shall be subject to the protection scope of the claims.

Claims

1. An intelligent information auditing method, characterized in that, The intelligent information auditing method comprises: S1, acquiring an information image set, classifying all information image pieces in the information image set, and obtaining a classification result; S2, according to the classification result, corresponding to a more matched information auditing process; S3, completing the information auditing process and obtaining an information auditing result. 2.The intelligent information auditing method of claim 1, wherein, The step S1 comprises: S11, acquiring an information image set, classifying all information image pieces in the information image set according to the recognizable degree, obtaining a conventional image set with a recognizable degree higher than a preset threshold, and obtaining a specific image set other than the conventional image set. 3.The intelligent information auditing method of claim 2, wherein, The step S2 comprises: S21, according to the classification result, corresponding to a first conventional information auditing process matched with the conventional image set; S22, according to the classification result, corresponding to a second special information auditing process with a higher operation complexity than the first conventional information auditing process. 4.The intelligent information auditing method of claim 3, wherein, The first conventional information auditing process comprises: Based on the cropped full convolution network text detection model, the conventional image set is converted into a first text detection feature map; The first text detection feature map is converted into a binary image of a text region by using a binary module, and the position information of a text box is obtained through a post-processing algorithm; The quantized deep separable convolution is used as a feature extraction network to extract a second text recognition feature map, the second text recognition feature map is input into a CTC layer for recognition, an optimal alignment path between the input feature sequence and the target text sequence is found, and finally a first predicted text sequence is obtained based on the optimal alignment path, thereby completing the auditing process. 5.The intelligent information auditing method of claim 4, wherein, The second special information auditing process comprises: The specific image set is subjected to bottom layer visual feature extraction, feature aggregation and dimension compression to obtain a first vector sequence; The uniform target dimension is set to a preset value, the first vector sequence is subjected to dimension conversion through a linear embedding layer to obtain a second vector sequence; the image feature and the text feature of the second vector sequence have consistent dimensions; The second vector sequence is added with position coding in a sine-cosine position coding manner to obtain a first predicted text sequence, thereby completing the auditing process. 6.The intelligent information auditing method of claim 5, wherein, The step S3 comprises: S61, acquiring N specific image pieces with a recognition confidence not greater than a preset threshold in the second special information auditing process as N image pieces to be processed; S62, calling a past first image piece recognition result, and judging whether there is an image piece with a similarity greater than the confidence in the past first image piece recognition result and the N image pieces to be processed; S63, if there is, calling the recognition result corresponding to the image piece with the similarity greater than the confidence as a first part of the information auditing result. 7.The intelligent information auditing method of claim 6, wherein, After the step S63, further comprising: If there is not, acquiring a manually input trusted recognition result, establishing a corresponding relationship between the image piece to be processed and the trusted recognition result, adding the corresponding relationship to the past first image piece recognition result to form a second part of the information auditing result, and combining the first part of the information auditing result to form the information auditing result.

8. An intelligent auditing system, characterized in that, The intelligent auditing system is used to implement the intelligent information auditing method as claimed in any one of claims 1 to 7. The intelligent auditing system is used to implement the intelligent information auditing method as claimed in any one of claims 1 to 7.

Images