Parallel multi-scale CNN for image denoising
Taoling Zhao, Yi Zhong, Yuanli Wang · 2019
Deep learning has achieved outstanding performance in the field of image processing. Inspired by the inception model, we propose a deep neural network convolving simultaneously at two different scales. We utilize the convolution kernel of different sizes to carry out the convolution, which can obtain the features of the image at different scales. The features of the output are no longer uniformly distributed, but converge the features with strong correlation. Batch normalization (BN) and residual learning (RL) are utilized to speed up the training process as well as enhance the denoising ability of model. The experimental results of three salt-and-pepper noise levels based on the public dataset BSDS500 show that the newly proposed method can obtain significantly superior performance than current state-of-the-art approaches and higher peak signal to Noise Ratio (PSNR) in salt-and-pepper noise removal.