General fusion method for infrared and visual images via latent low-rank representation and local non-subsampled shearlet transform

Abstract

This study establishes the general fusion method for infrared and visual images via latent low-rank representation (LatLRR) and local non-sampled shearlet transform (LNSST) to effectively combine the salient information of both images and solve problems on low-contrasting heterogeneous image fusion. First, LNSST is used as a multi-scale analysis tool to decompose the source images into low-pass and high-pass sub-images. Second, the LatLRR, which is an effective method for exploring multiple subspace structural data, is used to extract the salient information of image sources. Thus, the LatLRR can be adopted to guide the adaptive weighted fusion of low-pass sub-images. Simultaneously, the average gradient, which can reflect image edge details, is regarded as the fusion rule for high-pass sub-images. A series of images from diverse scenes are used for the fusion experiments, and the results are evaluated subjectively and objectively. The subjective and objective evaluations show that our algorithm exhibited superior visual performance, and the values of the objective evaluation parameters increase by about 5–10% compared with other typical fusion methods.