Research On Image Restoration Methods Based On Deep Learning

Iris recognition has become one of the most promising biometric technologies because of its advantages of high accuracy and non-contact.However,iris images often have problems such as insufficient resolution and easy blurring during the acquisition process,making the recognition accuracy of the existing iris recognition system not ideal.From the perspective of the practical application of an iris recognition system,this paper studies the iris image quality enhancement method given the problem of iris recognition performance degradation caused by low-quality images and the shortcomings of existing methods.The research results are as follows:First,the existing methods generally use convolutional neural networks,which can only use local information and not global information during the training process,resulting in weak feature extraction ability and inability to restore iris texture accurately.In this paper,an iris image quality enhancement me thod based on the channel attention mechanism is proposed,which can model long-distance pixel dependence,dynamically adjust parameters according to input during inference,and also realize linear computational complexity,making the calculation amount of this method smaller than the existing methods based on attention mechanism.Experimental results show that the proposed method can restore high-quality iris images and significantly improve the performance of iris recognition.Second,iris images have complex microstructures,making the iris restoration task more difficult.Existing deep learning-based methods typically generate the final result image directly without utilizing iris-specific prior information.In this paper,a generative adversarial network is pre-trained with iris images to obtain iris prior knowledge.This generative adversarial network can generate textured iris images,so the network weights contain rich iris prior knowledge.Next,a feature modulator embeds generative iris prior knowledge into the image restoration.Finally,the algorithm obtains high-quality iris images and can improve the recognition accuracy of the iris recognition system.