Multimodal image fusion via coupled feature learning

This paper presents a multimodal image fusion method using novel decomposition model based on coupled dictionary learning. The proposed is general and can be used for variety of imaging modalities. In particular, the images to fused are decomposed into correlated uncorrelated components sparse representations with identical supports Pearson correlation constraint, respectively. resulting optimization problem solved by an alternating minimization algorithm. Contrary other learning-based methods, approach does not require any training data, features extracted online from data itself. By preserving in images, significantly improves current approaches terms maintaining texture details modality-specific information. maximum-absolute-value rule only. leads enhanced contrast-resolution without causing intensity attenuation or loss important Experimental results show that achieves superior performance both visual objective evaluations compared state-of-the-art methods.