Image processing method and system based on sparse auto-encoder, medium and equipment

Abstract

The invention belongs to the technical field of image processing, and discloses an image processing method, system, medium and device based on a sparse auto-encoder, and the method comprises the stepsof presetting the deviation of all neural networks as a given vector or matrix, and carrying out the continuous updating; introducing a cognitive feedback weight matrix into the deviation, and improving the accuracy of the sparse auto-encoder network by utilizing the feedback deviation; controlling the sparse auto-encoder based on a regularization method of the H1 norm. The invention discloses adenoising sparse auto-encoder based on feedback deviation and H1 regularization. In the model, a feedback bias matrix is introduced, so that the bias of the network can be more accurately controlled.The feedback deviation changes with the number of iterations. H1 regularization can reduce overfitting and can also prevent more important features from being excessively smoothed. Experimental results show that the method is effective and is superior to the existing DnCNN-3, DAEP and EPCNN methods.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to an image processing method, system, medium, and device based on a sparse autoencoder. Background technique [0002] At present, in image processing, many images are inevitably not affected by noise, and some image processing methods appear accordingly. In image processing, image noise reduction is very important. Image denoising is still challenging at this stage due to image blurring and artifacts that occur during the denoising process. The development of noise reduction algorithms is a necessary task. In most models, the noise is assumed to be Gaussian or salt-and-pepper noise, and many algorithms are designed to remove the effects of reducing noise. [0003] So far, many traditional noise reduction algorithms have appeared, such as: anisotropic dissipation filter, non-local mean filter, BM3D, etc. However, most of these traditional image denoising alg...

AI Technical Summary

Problems solved by technology

[0004] Through the above analysis, the problems and defects of the existing technology are: in the process of denoising the image by the sparse autoencoder in the prior art, some important detail features will be lost

Images