Optimized Multimodal Neurological Image Fusion Based on Low-Rank Texture Prior Decomposition and Super-Pixel Segmentation

Abstract

Multimodal image fusion has become a crucial tool in the field of computer-assisted diagnosis due to its powerful ability to provide a concise and accurate visualization of important diagnostic information. This paper presents a multimodal neurological image fusion method using an optimized low-rank texture prior (OLTP) model and super-pixel segmentation. At first, the source images are decomposed into cartoon and texture components using the OLTP model. The texture components are fused using super-pixel segmentation based on a pixel-related Gaussian mixture model to preserve the local oscillating patterns without introducing edge distortion. To fuse cartoon components, a weighted energy fusion rule is used to preserve the global structural information. Grey wolf optimization is used to tune the regularization coefficients of OLTP model to balance the characterization of cartoon and texture components. The efficacy of the proposed method is demonstrated using substantial experiments conducted on various multimodal image pairs. The experimental results validate the superiority of the proposed method over the state-of-the-art fusion methods in both visual and objective performance.