Mixed-Noise Removal in Images Based on a Convolutional Neural Network

Aiming at limiting drawbacks of denoising algorithms that can only remove one or two specific types of noise (and which are inefficient for other types), we propose a combined neural-network model for mixed-noise removal in images. Nine convolutional layers are adapted, and noisy images are trained through feature extraction, shrinking, nonlinear mapping, expanding, and reconstruction. Experimental results show that the algorithm achieves better denoising results and is more suitable than other algorithms for dealing with different types of mixed noise in images. Subjective visual effects and an objective evaluation demonstrate the achieved improvements.