Research On Image Denoising Algorithm Based On Sparse Coding And Deep Learning

The image denoising problem is an important research direction in the image restoration work.This problem has been widely studied by scholars at home and abroad today because of the increasingly high requirements for image quality.The current mainstream image denoising algorithms are divided into model-based methods and learning mapping methods.The former is represented by a sparse representation model,mining the prior knowledge inherent in clean images.The latter uses the popular deep learning technology in recent years as a representative to complete the image feature extraction work,and input image data sets for network model training to obtain image detailed feature information.In order to take advantage of the two algorithms in image denoising tasks,this paper conducts research on the denoising problem of noisy images,using improved algorithms of sparse representation and deep learning to perform denoising,and simulations to finally obtain experimental data and analyze.The research work of this paper is as follows:(1)Learn the sparse representation theory and apply it to image denoising related algorithms,based on this,a sparse representation denoising model using kernel methods and feature distance matrices is proposed.The main idea is to use the training dictionary to sparsely represent a noisy image.Through the feature that the noise information cannot be sparsely represented by clean image blocks,the sparse representation vector and dictionary image block reconstruct the image without noise information,thereby achieving the effect of removing image noise.(2)A deep learning denoising model is designed.A residual algorithm and a batch normalization algorithm are added to the network layer that learns the image noise information features.The zero-filled convolution operation is used to ensure the consistency of the input and output feature dimensions of each layer.The output of the network model is image noise information.The model is divided into two residual convolutional network models.The output of the two sub-networks is the noise information in the image.The outputs are weighted and summed to obtain the fused noise information.The noise image is subtracted from the noise image to obtain the denoised image.(3)Generate a hybrid denoising model by weighting the two sub-models of sparse representation denoising and deep learning denoising,and determine the denoising result weight parameters of the two denoising models through multiple simulation experiments to obtain the best denoising.performance.