Document Type : Research Paper
PhD. Student of Lorestan University
Department of Electrical and Electronics Engineering, Faculty of Engineering, Lorestan University
Noise removal is one of the important topics in image processing to improve image quality. Since deep neural networks alone face problems such as vanishing gradient by increasing the depth of the network and could not cover the details well for a specific task, in this paper, by reducing the depth of the network and increasing its width, it is possible to obtain diverse features from different channels, which increases the accuracy of the network. By increasing the width of the network into two branches, different information is extracted from the noisy image, which is very accurate for separating noisy data from the image. The effect of attention to the features in each channel and their weighting in the noise removal operation is also considered. Finally, to check the effectiveness of the proposed method, the results are compared with the state of the art results in this field. The simulation results by examining both PSNR and SSIM and from a visual point of view show that this method is acceptable in dealing with various types of synthetic and blind noises for performing processing, and on the other hand, using the effect of attention and parallel networks can be achieved with PSNR=36.34db, which is very effective in real noise.