A multi-modal image fusion framework based on guided filter and sparse representation. (February 2021)

• Record Type: Journal Article
• Authors: Zhang, Shuai; Huang, Fuyu; Liu, Bingqi; Li, Gang; Chen, Yichao; Chen, Yudan; Zhou, Bing; Wu, Dongsheng
• Is Part Of: Optics and lasers in engineering. Volume 137(2021)
• Publication Date: 2021-02
• Subjects: Multi-modal image fusion -- Cartoon-texture decomposition -- Sparse representation -- Multi-scale guided filter -- Saliency map; Lasers in engineering -- Periodicals; Optical measurements -- Periodicals; Optics -- Periodicals; Lasers en ingénierie -- Périodiques; Mesures optiques -- Périodiques; Optique -- Périodiques; 621.36605
• DOI: 10.1016/j.optlaseng.2020.106354
• Languages: English
• ISSNs: 0143-8166

Abstract

Highlights: An overcomplete dictionary of cartoon components is constructed and it gives a rise to the robust and universal representation of the cartoon components. A modified OMP is proposed to enhance the efficiency of sparse coding. The iterative visual saliency map is proposed to compute the weight map. The multi-scale guided filter is then designed to optimize the weight map to extract the scale-aware structures and improve the spatial consistency. A multi-model image fusion framework is proposed based on cartoon-texture decomposition. On the basis of it, the structures are better integrated into the fused image. Abstract: To completely integrate the structures of images, a cartoon-texture decomposition based image fusion framework is proposed in this paper. The cartoon components represent the background and main structures information. Due to its high similarity, the sparse representation is utilized based on an overcomplete cartoon dictionary. Moreover, a modified orthogonal matching pursuit is proposed to improve the fusion image continuity and eliminate artifacts interferences. The texture components describe more distinct structures, which have diverse scale characteristics. To preserve more relevant infrared details in fused texture, an iterative visual saliency map is proposed. Furthermore, the weight maps are computed on the basis of a designed multi-scale guided filter and saliency map; then, the weight maps are utilized to guide the texture fusion, aiming to improve the visual and scale-aware salience. The final fused image is obtained by combining the cartoon fusion and texture fusion results. The comprehensive and objection evaluations demonstrate the outperformance on accuracy, robustness and versatility of the proposed fusion framework.