Image Zooming Models Based On Image Decomposition And Convolutional Neural Network

The traditional interpolation amplification method is not easy because the continuity between pixels is not strong.Causes magnified image edges with jagged and blurred images.In recent years,partial differential equations have been advancing and developing in the field of image processing theory and practical applications.More and more researchers have applied partial differential equations to image processing and the application of deep learning techniques in images has become widespread.In this paper,partial differential equations and deep learning are used as image processing tools to introduce image decomposition,image diffusion and convolution neural networks into image zooming.The image decomposition model and image diffusion model based on partial differential are studied respectively,an image zooming model based on image decomposition and partial differential equation is proposed.Super-resolution Reconstruction of CNN Images Based on Fusion Multiscale Feature Information.The main work of this paper are as follows:First,this paper proposes a partial differential image zooming method based on image decomposition combining image decomposition model and partial differential diffusion model.According to the image decomposition theory,the image can be decomposed into a cartoon part and a texture part,aiming at the difference between these two parts.The composition uses the second-order isotropic and an-isotropic diffusion model respectively.At the same time,to improve the zooming effect,the fourth-order an-isotropic partial differential equation is coupled with the second-order diffusion model as the diffusion model.Simulation experiments show that the proposed lethod can maintain the image edge and image clarity.The objective evaluation criteria also prove the effectiveness of this method.Second,combining convolution neural network and feature pyralid structure,Super-resolution Reconstruction of CNN Images Based on Fusion Multiscale Feature Information,which fused the low-level features and high-level feature information of the image.Experiments have verified the effectiveness of the improved method.