CN-121982740-A - Image processing method, device, equipment, medium and product

CN121982740ACN 121982740 ACN121982740 ACN 121982740ACN-121982740-A

Abstract

The invention discloses an image processing method, an image processing device, image processing equipment, medium and a product, relates to the field of image processing, relates to the field of artificial intelligence, in particular to application of a large model in the field of financial technology, can be used in the field of financial technology, acquires a bill image of a target bill, processes the bill image to obtain a standard image, detects and classifies the standard image based on a multi-mode large model and a decision rule base to obtain a processing type of the standard image and processing information corresponding to the processing type, and processes the standard image according to the processing type and the processing information. By adopting the technical scheme, the images are classified by combining the multi-mode large model and the decision rule base, the images are processed based on the processing types, intelligent filtering of the image quality is realized, the accuracy and the efficiency of image processing are improved, and the problems of resource waste caused by blind receiving and blind repairing of the image processing are solved.

Inventors

YANG RISHENG

Assignees

中国工商银行股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260129

Claims (11)

1. An image processing method, comprising: acquiring a bill image of a target bill; Processing the bill image to obtain a standard image; detecting and classifying the standard image based on the multi-mode large model and a decision rule base to obtain a processing type of the standard image and processing information corresponding to the processing type, wherein the processing type comprises identification processing, repair processing and re-shooting processing; And processing the standard image according to the processing type and the processing information.
2. The method according to claim 1, wherein the detecting and classifying the standard image based on the multi-mode large model and the decision rule base to obtain the processing type of the standard image and the processing information corresponding to the processing type includes: Carrying out multi-dimensional detection on the standard image based on the multi-mode large model and prompt words corresponding to the multi-mode large model to obtain multi-dimensional information, multi-dimensional detection values and image detection values, wherein the multi-mode large model comprises a light detection model, a fuzzy detection model and an integrity detection model; comparing the image detection value with a classification threshold value to obtain a detection type; Determining detection feedback from an anomaly detection feedback library based on the multi-dimensional detection value and the detection type; formatting the multi-dimensional information, the multi-dimensional detection value, the detection type and the detection feedback to obtain a detection result; carrying out decision judgment on the detection result based on a decision rule base to obtain a decision result, wherein the decision result comprises a decision type and decision feedback; Determining a processing type of the standard image based on the detection type and the decision type; and integrating the decision feedback, the detection feedback and the multidimensional information based on the processing type to obtain processing information corresponding to the processing type.
3. The method of claim 2, wherein the cue words comprise a light cue word, an ambiguity cue word, and an integrity cue word, and wherein the cue words are constructed in the following manner, respectively: obtaining contrast factors, exposure factors and shadow factors by carrying out light detection on an image to be processed, and carrying out weighting treatment on the contrast factors, the exposure factors and the shadow factors to determine light detection values; recording a quadrant shadow position of the image to be processed based on the shadow factors, and outputting the light detection value and the quadrant shadow position as light for light detection; constructing a light prompt word according to a light detection mode and the light output; detecting a key region of the image to be processed to obtain at least one region fuzzy value, and determining a fuzzy region from the key region based on the region fuzzy value and a fuzzy threshold; outputting the region fuzzy value and the fuzzy region as fuzzy output; constructing a fuzzy prompt word according to a fuzzy detection mode and the fuzzy output; calculating the four-corner missing ratio of the image to be processed as an integrity detection value, and outputting the integrity detection value as the integrity of the integrity detection; and constructing an integrity prompt word according to the integrity detection mode and the integrity output.
4. The method of claim 3, wherein the multi-dimensional detection values include a light detection value, a blur detection value, and an integrity detection value, and wherein the multi-dimensional detection of the standard image based on the multi-modal large model and the hint words corresponding to the multi-modal large model to obtain multi-dimensional information, multi-dimensional detection values, and image detection values comprises: Performing light detection on the standard image based on a light detection model in the multi-mode large model and the light prompt word to obtain a light detection value and shadow quadrant information; performing fuzzy detection on the standard image based on a fuzzy detection model in the multi-modal large model and the fuzzy prompt word to obtain fuzzy region information and a fuzzy detection value; carrying out integrity detection on the standard image based on the integrity detection model in the multi-modal large model and the integrity prompt word to obtain an integrity detection value; And carrying out weighted aggregation on the light detection value, the fuzzy detection value and the integrity detection value to obtain an image detection value.
5. The method according to claim 2, wherein the decision-making of the detection result based on the decision rule base, to obtain a decision result, comprises: determining an abnormal region of the standard image according to the multi-dimensional information in the detection result; determining decision feedback from a feedback association relationship of a decision rule base according to the abnormal region and the multi-dimensional detection value in the detection result; Integrating the multi-dimensional detection values to obtain decision detection values; comparing the decision detection values based on a decision threshold to obtain a decision type; and taking the decision type and the decision feedback as decision results.
6. The method of claim 2, wherein the determining the processing type of the standard image based on the detection type and the decision type comprises: if the detection type is inconsistent with the decision type, the decision type is used as the processing type of the standard image; determining decision feedback corresponding to the decision type; And optimizing the multi-mode large model by the decision feedback and the standard image.
7. The method according to claim 2, wherein integrating the decision feedback, the detection feedback and the multi-dimensional information based on the processing type to obtain processing information corresponding to the processing type comprises: If the processing type is repair processing, determining a repair type based on the detection feedback and the decision feedback; And determining a repair location based on the multi-dimensional information; The repair type and the repair position are used as the corresponding processing information of the repair processing; if the processing type is the re-shooting processing, comparing the multi-dimensional information based on a multi-dimensional abnormal threshold value to obtain the re-shooting type; Filling the multi-dimensional information into a re-shooting suggestion template corresponding to the re-shooting type to obtain a re-shooting suggestion; and recommending the re-shooting proposal as processing information corresponding to the re-shooting processing.
8. The method of claim 1, wherein processing the standard image according to the processing type and the processing information comprises: If the processing type is identification processing, directly carrying out text identification on the standard image to obtain a bill identification result; If the processing type is repair processing, determining a target repair processor from at least one candidate repair processor in a repair scheme library based on the repair type in the processing information; Repairing the standard image based on the target repairing processor and the repairing position in the processing information to obtain a repairing image; Text recognition is carried out on the repair image, so that a bill recognition result is obtained; And if the processing type is the re-shooting processing, displaying the processing information of the processing type through an interactive interface based on a fixed format.
9. A transaction abnormality detection device, comprising: the acquisition module is used for acquiring the bill image of the target bill; the preprocessing module is used for preprocessing the bill image to obtain a standard image; The detection classification module is used for detecting and classifying the standard image based on the multi-mode large model and the decision rule base to obtain the processing type of the standard image and the processing information corresponding to the processing type, wherein the processing type comprises identification processing, repair processing and re-shooting processing; and the processing module is used for processing the standard image according to the processing type and the processing information.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores computer instructions for causing a processor to implement the image processing method of any one of claims 1-8 when executed.
11. A computer program product, characterized in that the computer program product comprises a computer program which, when executed by a processor, implements the image processing method according to any of claims 1-8.

Description

Image processing method, device, equipment, medium and product Technical Field The invention relates to the field of image processing, relates to the field of artificial intelligence, in particular to application of a large model in the field of financial science and technology, and can be used in the field of financial science and technology, in particular to a transaction abnormality detection method, device, equipment, medium and product. Background The processing of bank notes is a core link of financial business, and relates to key processes of note information input, authenticity verification, payment settlement and the like. With the development of finance technology, the image digital processing of paper bill has become industry standard, which covers the steps of scanning and uploading, OCR (Optical Character Recognition, optical character) character recognition, seal detection, anti-counterfeiting verification and the like. For the photos uploaded by users, the current mainstream scheme mostly adopts fixed pipeline processing of 'repair before identification', and most of the current mainstream scheme adopts traditional algorithms such as perspective correction, histogram equalization and other methods for preprocessing, then adopts the traditional scheme for image repair, and finally adopts some traditional algorithms and deep learning algorithms for OCR and seal detection. However, invalid restoration of low-quality images consumes more than 60% of calculation power, and lacks a user guidance mechanism, so that the repeated submission rate is too high, when the process forcibly restores an image which is severely damaged, information distortion and nonsensical calculation power waste are caused, and when an erroneous image is subjected to OCR recognition, the false recognition rate is improved. Disclosure of Invention The invention provides an image processing method, an image processing device, image processing equipment, an image processing medium and an image processing product, which solve the problems of resource waste caused by blind receiving and blind repairing in the image processing process, classify images by combining a multi-mode large model and a decision rule base, process the images based on processing types, realize intelligent filtering of the images based on image quality and improve the accuracy and efficiency of image processing. According to an aspect of the present invention, there is provided an image processing method including: acquiring a bill image of a target bill; Processing the bill image to obtain a standard image; detecting and classifying the standard image based on the multi-mode large model and a decision rule base to obtain a processing type of the standard image and processing information corresponding to the processing type, wherein the processing type comprises identification processing, repair processing and re-shooting processing; And processing the standard image according to the processing type and the processing information. According to another aspect of the present invention, there is provided a transaction abnormality detection device including: the acquisition module is used for acquiring the bill image of the target bill; the preprocessing module is used for preprocessing the bill image to obtain a standard image; The detection classification module is used for detecting and classifying the standard image based on the multi-mode large model and the decision rule base to obtain the processing type of the standard image and the processing information corresponding to the processing type, wherein the processing type comprises identification processing, repair processing and re-shooting processing; and the processing module is used for processing the standard image according to the processing type and the processing information. According to another aspect of the present invention, there is provided an electronic apparatus including: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the image processing method according to any one of the embodiments of the present invention. According to another aspect of the present invention, there is provided a computer readable storage medium storing computer instructions for causing a processor to execute the image processing method according to any one of the embodiments of the present invention. According to another aspect of the present invention, there is provided a computer program product comprising a computer program which, when executed by a processor, implements an image processing method according to any of the embodiments of the present invention. According to the technical scheme, the images are classified by combining the multi-mode large model and the decision rule base, the images are processed based on the processing types