Image Fusion Based on Gradient Regularized Convolution Sparse Representation

Abstract

An image fusion method based on gradient regularized convolution sparse representation is proposed, which makes up for the shortcoming of conventional method. Target image is composed of optimal high frequency and low frequency by two scale decomposition of source image with sparse optimization function. The high frequency components are obtained by convolution sparse representation model and alternative direction multiplier method, which could raise ability to maintain image details, and low sensitivity to image registration. Optimal low frequency components are obtained with the strategy of maximum or average. Experimental results demonstrate that proposed method has a great improvement in details preserve of image.