Research On Key Technologies Of Hand Vein Information Image Inpainting

Image inpainting technology is to predict the missing part of some damaged and missing images according to the known content information.Although image restoration technology is widely used at present,there is little work on the restoration of hand vein feature information.With the research and application of hand vein feature information recognition becoming more and more extensive,the integrity of hand vein image information has higher requirements.Therefore,how to complete the damaged hand vein image information needs to be studied at present direction.At present,the traditional digital image restoration technology is based on manual annotation,which makes it difficult to select the feature information.When the missing area is large,the ability to repair the image structure information and high-level texture information will be greatly reduced.In recent years,with the excellent performance of the deep neural network in image feature learning,the research of image restoration based on the deep neural network has gradually become a hot spot.However,the current image restoration algorithm still has instability and the low quality of the repaired image.Therefore,because of the above problems,this paper puts forward the repair algorithm of hand vein information from three directions,respectively,from the aspects of deep convolution neural network,generation of confrontation network,image and image conversion and so on.Finally,through the comparison and evaluation with the related direction algorithm,the repair effect of hand vein image information has been improved to some extent.(1)In the basic u-net network,a space conversion network layer is introduced by leaping,that is,the deformation convolution network module,which is used to repair the whole structure and detail information of the missing image,to better learn the relationship between the missing information and other image information around.To reduce the loss between the decoder function and the real sample,the loss of the U-net network decoder function is introduced.Under this constraint,the decoder features in the missing region can be used to guide the movement of the encoder features in the known region,to improve the quality of the repaired image.(2)According to the distribution characteristics of the vein image information of the dorsum of the hand,a cascade depth generation network restoration framework is used to fuse the vein segmentation image and the vein image,and effectively use the key information of the vein image.To effectively mine the geometric information features of the hand vein image and the segmented image,the nonlocal network block is introduced into the repair network.The global and local confrontation loss and image perception loss are used to ensure the consistency between the whole and detail of the vein image and the original image.(3)Based on the work of image to image conversion,data is decomposed into a shared part and a unique part through the content network and feature network to achieve disentangled and representation.At the same time,based on the representation learning model of mutual information estimation,we learn from the key points of the vein and the complete image in pairs,to learn about the separation representation of the vein network from the point production line.Based on the resistance loss and the perceived loss,we add the loss of cycle consistency,to realize the good repair of the seriously missing vein image through the key points.The hand vein image data set used in this paper is a self-made hand vein image collected in the laboratory.During the training process,the image data set of hand veins is preprocessed according to the needs.Experimental results show that the algorithm proposed in this paper has a certain improvement in the repair effect and better performance in the objective evaluation index.There are 31 figures,12 tables and 84 references in this paper.