Research On Image Super-Resolution Reconstruction Based On Convolution Neural Network And Residual Dictionary Learning

Image super resolution reconstruction is to reconstruct high resolution images based on one or more low resolution images.High resolution images usually contain more abundant information.It also provides a good research foundation for other research directions in image processing,while meeting the human needs for visual effects.However,due to the complexity of image degradation model,the problem of image super resolution reconstruction is usually an ill posed problem.Convolution neural network has the ability to fit arbitrary complex functions,and it can solve the ill posed problem of image super resolution reconstruction to a certain extent.However,there is still residual error between the reconstructed result and the target high-resolution image,which is usually high frequency information.In order to compensate the residual to improve the effect of image super resolution reconstruction,a new image super-resolution reconstruction algorithm based on convolution neural network and residual dictionary is proposed in this paper,which can make up the high and low resolution information loss between high and low resolution images while retaining structured information.First,on the basis of super-resolution convolution neural network,by improving the number of network layers,convolution kernel and the number of convolution kernel,the training parameters in the convolution neural network can be increased,which can better fit the image super-resolution reconstruction model.It improves the selection of the excitation function and uses PReLU instead of ReLU to increase its nonlinear mapping ability.Secondly,an end-to-end image super resolution reconstruction model is obtained by using the improved network in the training set.Thirdly,according to the case of high resolution image and high resolution image between the reconstructed image and the target high resolution image,the dictionary learning method is selected for the high frequency information sensitive dictionary,and the residual dictionary between the target high resolution image and the output image of the super-resolution convolution neural network is learned.Finally,the residual dictionary is used to compensate the missing high frequency information in the reconstructed image,so that the high resolution image with super resolution reconstruction effect can be obtained.Through contrast experiments,the algorithm is compared with other representative algorithms in subjective and objective image quality evaluation and time efficiency.Experimental results show that the proposed algorithm is better than the contrast algorithm in subjective image quality evaluation.On the subjective quality evaluation,the reconstructed image is clearer,and the details are recovered well.The peak signal to noise ratio and structural similarity are improved on the objective quality evaluation,but the time efficiency is slightly reduced.Therefore,the improved super-resolution convolution neural network can learn the mapping relation between effective high and low resolution images,and the optimization method of residual dictionary learning can compensate the high frequency information that can not be recovered in the super-resolution convolution neural network.The paper has 14 figures,8 tables and 52 references.