Research On Key Technologies Of Identity Document Recognition And Design Of Software System Based On Android

The identity document recognition system based on computer vision can capture and process the multi-spectral images of the document,automatically read the information of the document,and is widely used in airports,stations,hotels and other occasions.However,most of the identity document recognition software systems run on PC with poor portability.Besides,it's difficult to accurately locate the position of document in the image with complex background,and the color in the UV image of passport is usually distorted.To solve the above problems,we carried out the following research.Firstly,aiming at the problem of document location in complex background,we proposed a target location algorithm based on the fusion of lightweight edge detection network and wavelet edge detection.The bi-directional cascading network BDCN was transformed into a lightweight network,and the edge image obtained by the network was merged with the image obtained by wavelet edge detection,which not only filtered out most of the background interference but also retained the fine and accurate edge of the document.Then the coordinates of the corner points were obtained and the position of document was located accurately using the LSD algorithm.Secondly,for the problem of color distortion in the UV image of passport,we used a color transfer algorithm based on dense semantics to achieve the effect of color correction.Through a neural network we extracted the multilevel semantic features of the reference image and the source image,and the linear model of color correction was established.When solving the coefficients of linear model,we optimized the initialization condition through the improved Reinhard's color transfer algorithm,and reduced the number of iterations.Finally,we designed and implemented an identity document recognition software system based on Android,which is able to capture and process the images of document,recognize the machine readable code and read out the information inside the chip.The experimental results indicated that the proposed lightweight edge detection network has high accuracy and efficiency,and the document location algorithm can overcome different environmental interference.The color correction algorithm can effectively solve the problem of color distortion in the UV image of passport,and the result of correction is better than other color transfer algorithms.The identity document recognition software system has been successfully applied to an identity verification product of a domestic bank,which demonstrated the reliability and stability of the proposed system.