Research On Image Motion Blur Removal Algorithm Based On Blind Deconvolutio

Image deblurring is a classic task in low-level computer vision,which has attracted great attention from the image processing and computer vision community.The purpose of image deblurring is to recover a sharp image from a blurred input image.With the rapid development of camera equipment,image quality issues have attracted more and more attention.At present,the movement of camera equipment causes blurred pictures to become one of the most prominent problems of mobile camera equipment.Blind deconvolution(blind deconvolution)is also called blind deconvolution.Image deblurring based on blind deconvolution refers to recovering clear images under uncertain conditions of blurring.In recent years,image blind deblurring algorithms have received extensive attention and made great progress.Among them,the neural network(Neural Network,NN)is becoming more and more popular in sharpening blurred images,especially in the field of single-frame image deblurring,and has made remarkable progress in recent years.However,this field still faces many challenges.The main research content of this paper is as follows:(1)This dissertation introduces the development process of traditional deblurring algorithms and image deblurring algorithms based on deep learning.We propose a new network Deblur GAN-SE for single-frame image deblurring,which is also an "end-to-end" Generative Adversarial Networks(GAN).With the participation of this network model,the defuzzification efficiency,quality and flexibility of the algorithm can be effectively improved.The Deblur GAN-SE network is a conditional GAN with a dual-scale discriminator,and this dissertation adds channel attention to its generator to increase the receptive field of the model,and the obtained image processing results are closer to realistic clear images.(2)In order to verify the possibility of the algorithm to achieve motion blur in video images,this paper proposes a deblurring algorithm framework combined with a lightweight model.In order to be used for real-time video image deblurring,this paper studies a variety of lightweight network models.Finally,it is proved that Deblur GAN-SE can achieve very good deblurring quality and short running time in several benchmarks on the classic evaluation indicators.Furthermore,this paper verifies that the architecture is also effective for general image restoration tasks.(3)This dissertation uses the target detection framework SSD to detect the deblurred image to verify the performance improvement of the deblurring algorithm to the downstream algorithm.First,use the deblurring algorithm proposed in this paper to process the blurred image,and then use the SSD target detection framework to detect the processed image and the blurred image respectively.Finally,the effect of the deblurring algorithm is evaluated according to the probability of detecting the target.