Application Of Residual Auto-Encoder In Image Noise Reduction

Under the background of the rapid development of information nowadays,with the popularity of various digital products,image has become the most important information carrier in daily life.It contains a lot of information,and which is the main way for us to obtain external information.However,in the process of imaging and transmission,image quality is often impaired due to the influence of external environment and imaging equipment,which seriously affects the subsequent image processing and analysis.Therefore,image denoising has long been an important research branch in the field of machine vision.It guarantees the integrity of image information and removes as much useless information as possible.However,because the mathematical statistical distribution of image noise is the same in different practical applications,it is impossible for us to put forward a denoising algorithm once and for all to deal with all the noise.With the development of neural network technology,the denoising model based on deep neural network has shown good denoising ability,but this model often requires a large number of labeled data and clusters with superior performance to train the model.However,for some specific areas,such as seismic survey,Bio-image information and so on,it is almost impossible to obtain a large number of effective label images.In this paper,a sparse denoising self-encoder model based on residual regularization is proposed,which combines the current denoising self-coding neural network and the traditional sparse representation theory.Based on the reconstruction error minimization,increase regular correlation between the adjacent minimum residual block,so that the residual statistical properties is more close to the noise,to achieve a better denoising effect.In addition,the original image is cut into a uniform size image block by sliding window,so that the new method can train the self-encoder network with only one picture and complete the task of image denoising.The noise reduction model proposed in this paper is tested on different image data sets.The PSNR and SSIM values of the image after noise reduction are analyzed qualitatively and quantitatively,and compared with the classical noise reduction algorithm.The experimental results show that the sparse self-encoder model based on residual denoising has good denoising performance and can directly denoise the original image.It has very high application value.