Search

CN-121545025-B - External license anti-counterfeiting feature detection method based on convolutional neural network model

CN121545025BCN 121545025 BCN121545025 BCN 121545025BCN-121545025-B

Abstract

The invention relates to the technical field of electric digital data processing, in particular to an external license anti-counterfeiting feature detection method based on a convolutional neural network model, which comprises the steps of inputting a license surface image into a target convolutional layer in a preset convolutional neural network to obtain a target output image; the method comprises the steps of determining an effective information area in a target output image by utilizing a gradient amplitude sequence of a target convolution layer, determining information effective parameters of a current extraction area corresponding to any pixel point in the effective information area and each gradient amplitude in the current extraction area, determining the current extraction area as input to obtain a mapping extraction area and an actual extraction area in a corresponding output image, and determining the authenticity result of an external license by utilizing pixel position characteristics and respective information effective parameters corresponding to the mapping extraction area and the actual extraction area. By the technical scheme, the authenticity of the external license can be more accurately identified.

Inventors

  • WEI YANLI
  • HAN CHUNTAO

Assignees

  • 德迅科技有限公司

Dates

Publication Date
20260508
Application Date
20251127

Claims (7)

  1. 1. An external license anti-counterfeiting characteristic detection method based on a convolutional neural network model is characterized by comprising the following steps of: obtaining a license surface image of an external license, and inputting the license surface image into a target convolution layer in a preset convolution neural network to obtain a target output image; determining an effective information area in a target output graph by utilizing a gradient amplitude sequence formed by pixel point gradient sum in each convolution kernel of a target convolution layer; determining a current extraction area corresponding to any pixel point in the effective information area, and determining information effective parameters by utilizing each gradient amplitude in the current extraction area; Determining a current extraction area as input to obtain a mapping extraction area and an actual extraction area in a corresponding output image, and determining pixel position characteristics corresponding to the mapping extraction area and the actual extraction area respectively; Determining the authenticity result of the external license by using the pixel position characteristics and the respective information effective parameters corresponding to the mapping extraction area and the actual extraction area; The determining the effective information area in the target output graph by using the gradient amplitude sequence formed by the pixel point gradient sum in each convolution kernel of the target convolution layer comprises the following steps: Determining an amplitude difference sequence corresponding to the gradient amplitude sequence by utilizing a gradient amplitude sequence formed by pixel point gradient sum in each convolution kernel of the target convolution layer; Determining an effective information area in the target output graph by utilizing the position of the maximum difference element in the amplitude difference sequence; the determining the effective information area in the target output graph by using the maximum difference element position in the amplitude difference sequence comprises the following steps: Determining two adjacent amplitudes corresponding to the maximum difference element position in the amplitude difference sequence, wherein the large amplitude in the adjacent amplitudes is used as a right end point of the sequence in the gradient amplitude sequence; taking the position of the maximum amplitude element in the gradient amplitude sequence as a sequence left end point, and intercepting from the gradient amplitude sequence based on the sequence left end point and the sequence right end point to obtain an analysis amplitude region; The analysis magnitude region is marked in the target output graph to determine a valid information region in the target output graph.
  2. 2. The method for detecting the anti-counterfeit feature of the external license based on the convolutional neural network model according to claim 1, wherein the determining the current extraction area corresponding to any pixel point in the effective information area comprises the following steps: and combining any pixel point in the effective information area with surrounding pixel points with the same gradient direction to obtain a current extraction area corresponding to the any pixel point.
  3. 3. The method for detecting the anti-counterfeit feature of the external license based on the convolutional neural network model according to claim 1, wherein the determining the information effective parameter by using each gradient amplitude in the current extraction area comprises the following steps: and determining the information effective parameters of the current extraction area by using the morphological characteristics, the average gradient amplitude, the standard deviation of the amplitude and the gradient amplitude of any pixel point of the current extraction area.
  4. 4. The method for detecting the anti-counterfeit feature of the external license based on the convolutional neural network model according to claim 3, wherein the determining the information effective parameter of the current extraction area by using the morphological feature, the average gradient amplitude, the standard deviation of the amplitude and the gradient amplitude of any pixel point of the current extraction area comprises the following steps: determining an external rectangle of a current extraction area in a target output graph, and determining the length-width ratio of the external rectangle; and determining the information effective parameters of the current extraction area by utilizing the aspect ratio, average gradient amplitude, amplitude standard deviation and gradient amplitude of any pixel point of the circumscribed rectangle of the current extraction area.
  5. 5. The method for detecting the anti-counterfeit feature of the external license based on the convolutional neural network model according to claim 1, wherein the step of determining the current extraction region as the input to obtain the mapped extraction region and the actual extraction region in the corresponding output image comprises the steps of: the current extraction area is used as input of a convolution layer to obtain a corresponding output image, and a corresponding mapping position area of the current extraction area in the corresponding output image is determined and used as a mapping extraction area; and determining an actual extraction area in the corresponding output image based on the effective information area in the corresponding output image of the current extraction area.
  6. 6. The method for detecting the anti-counterfeit feature of the external license based on the convolutional neural network model according to claim 1, wherein the determining the pixel position features corresponding to the mapped extraction region and the actual extraction region respectively comprises: Determining respective signal values and coordinate values of corresponding pixel points in the mapping extraction area and the actual extraction area; And taking the signal value and the coordinate value as pixel position characteristics corresponding to the mapping extraction area and the actual extraction area respectively.
  7. 7. The method for detecting the anti-counterfeit feature of the external license based on the convolutional neural network model as set forth in claim 6, wherein the determining the true or false result of the external license by using the pixel position features and the respective information effective parameters corresponding to the respective mapped extraction areas and the actual extraction areas comprises: determining signal difference values and point distances between pixel points corresponding to the mapping extraction area and the actual extraction area based on the signal values and the coordinate values; and determining the true and false result of the external license by using the information effective parameters of the mapping extraction area and the actual extraction area, the signal difference value and the point distance.

Description

External license anti-counterfeiting feature detection method based on convolutional neural network model Technical Field The invention relates to the technical field of electric digital data processing, in particular to an external license anti-counterfeiting feature detection method based on a convolutional neural network model. Background The external license needs to be anti-fake, so that dangerous molecules are prevented from bypassing the inspection to implement dangerous activities through the fake license, and meanwhile, the situation that the identity is faked due to the fake license is avoided, and the benefit of a stolen person is damaged. And along with the popularization of high-definition printing, digital image technology and the like, the cost of forging the license is reduced, and the anti-counterfeiting difficulty is increased. In the prior art, automatic identification of anti-counterfeiting information is performed through a Convolutional Neural Network (CNN), so that multi-scale perception of element information is performed through different depth layers in the convolutional layer, the variability and the concealment of anti-counterfeiting characteristics in an external license are adapted, an original image is input to the output of true and false judgment, the CNN can optimize the integrated process of characteristic extraction and classification through training, and the detection efficiency is improved. However, in the actual inspection process, the anti-counterfeiting feature area of the certificate is worn along with use, the imaging quality is poor, the imaging angle is deviated, and the like, so that deviation occurs in the photographed certificate image, and the accuracy of identifying the anti-counterfeiting area part of the certificate in the image is insufficient. Disclosure of Invention In order to solve the technical problem of low accuracy of identification of a license anti-counterfeiting area part in an external license image, the invention aims to provide an external license anti-counterfeiting feature detection method based on a convolutional neural network model, and the adopted technical scheme is as follows: the invention provides an external license anti-counterfeiting characteristic detection method based on a convolutional neural network model, which comprises the following steps: obtaining a license surface image of an external license, and inputting the license surface image into a target convolution layer in a preset convolution neural network to obtain a target output image; determining an effective information area in a target output graph by utilizing a gradient amplitude sequence formed by pixel point gradient sum in each convolution kernel of a target convolution layer; determining a current extraction area corresponding to any pixel point in the effective information area, and determining information effective parameters by utilizing each gradient amplitude in the current extraction area; Determining a current extraction area as input to obtain a mapping extraction area and an actual extraction area in a corresponding output image, and determining pixel position characteristics corresponding to the mapping extraction area and the actual extraction area respectively; And determining the authenticity result of the external license by using the pixel position characteristics and the respective information effective parameters corresponding to the mapping extraction area and the actual extraction area. Further, the determining the effective information area in the target output graph by using the gradient amplitude sequence formed by the pixel point gradient sum in each convolution kernel of the target convolution layer includes: Determining an amplitude difference sequence corresponding to the gradient amplitude sequence by utilizing a gradient amplitude sequence formed by pixel point gradient sum in each convolution kernel of the target convolution layer; and determining an effective information area in the target output graph by using the position of the maximum difference element in the amplitude difference sequence. Further, the determining the effective information area in the target output graph by using the maximum differential element position in the amplitude differential sequence includes: Determining two adjacent amplitudes corresponding to the maximum difference element position in the amplitude difference sequence, wherein the large amplitude in the adjacent amplitudes is used as a right end point of the sequence in the gradient amplitude sequence; taking the position of the maximum amplitude element in the gradient amplitude sequence as a sequence left end point, and intercepting from the gradient amplitude sequence based on the sequence left end point and the sequence right end point to obtain an analysis amplitude region; The analysis magnitude region is marked in the target output graph to determine a valid information region in the targe