Deep CNN with Residual learning and Dilated Convolution for Image Denoising
B. Ramaswamy Karthikeyan, Mokkala Mounika, Sirigireddy Buchireddy Pravallika, Karakala Mahitha · 2021 Second International Conference on Electronics and Sustainable Communication Systems (ICESC) · 2021
In the field of digital image processing, filtering noise from the image to extract a high-quality image is an important pre-processing step for object detection, segmentation, tracking, etc. Convolutional Neural Network (CNN) has gained great attention in the domain of Image Denoising, with its flexible architecture. There are some challenges in using the deep CNNs for image denoising tasks like training the model and performance saturation when the depth of the network is increased. In the proposed model, two networks are connected parallelly to increase the width of the network rather than depth, and hence obtain more context information and make it less prone to performance saturation. Batch Normalization improves the performance of the network and mitigates the internal covariate shift problem. Dilated convolutions are used in the proposed model to extract more features by enlarging the receptive field. Residual learning is adopted to overcome exploding gradient and vanishing gradient problems. Experimental results have shown that our proposed model is more efficient than many existing filters.