Research On Super-resolution Reconstruction Technology Of Finger Vein Based On Learning

Finger vein feature recognition technology has gradually become a research hotspot in the field of biometric recognition by virtue of its security,easy access,uniqueness and other advantages.However,limited by the shooting environment,hardware system,and hardware cost,the resolution of the obtained image often fails to reach the expected goal,and the low-resolution image will seriously affect the recognition rate of the recognition system.High-resolution images have many image details,and the detailed information carries rich image features,which is crucial for the recognition rate of finger vein images.In this case,it is a feasible solution to convert low-resolution images into high-resolution images through learning-based algorithms.Image super-resolution reconstruction can effectively convert low-resolution images into high-resolution images,thereby improving the recognition rate of the recognition system.Aiming at image super-resolution reconstruction technology,this thesis has done the following work:(1)The neighborhood embedding super-resolution reconstruction algorithm is one of the most effective algorithms in the super-resolution reconstruction algorithm based on learning,but this method simply uses the KNN nearest neighbor search to find the neighborhood set of the test image during image reconstruction.The training set is not fully utilized.When there is no image very similar to the test image in the training set,the result is only the K images with the smallest Euclidean distance from the test image found in the existing training set,which leads to the possibility of incomplete reconstruction of high-resolution images.Satisfactory.To solve this problem,this thesis proposes an improved neighborhood-embedding algorithm,which uses neighborhood reconstruction algorithm in the neighborhood-embedding algorithm,which can reconstruct the neighborhood set of the test image and find the neighborhood image that is more similar to the test image.,Thereby improving the performance of reconstruction.Simulation experiments show that the algorithm in this thesis improves the performance of image reconstruction,and has a certain improvement in the recognition rate of finger veins.(2)The reconstruction performance of the neighborhood embedding super-resolution reconstruction technology is very good,but this method has a large number of matrix operations and the training and reconstruction phases are inseparable.These factors cause its time complexity to be too high.Therefore,this thesis studies the super-resolution reconstruction method based on deep learning.Most of the existing methods are based on residual networks,and only use shallow and deep features for super-resolution reconstruction,but do not make full use of residuals.The local feature information of the network is used for image reconstruction.In response to the above problems,this thesis proposes an improved WDSR(Wide Activation Super-Resolution)algorithm for superresolution reconstruction of finger vein images.The algorithm can adaptively use the local feature information of the deep neural network to perform super-resolution reconstruction of finger vein images.Simulation experiments show that the algorithm proposed in this thesis has a certain improvement in image reconstruction performance than the existing algorithms,and at the same time has a significant improvement in time complexity compared to the neighborhood-embedded superresolution reconstruction algorithm.