Infrared and Visible Image Fusion Based on Latent Low-Rank Representation

Infrared and visible image fusion is widely used in target tracking, detection, and recognition. To preserve image details and enhance contrast, this study proposed an infrared and visible image fusion method based on latent low-rank representation. The latent low-rank representation was used to decompose the source images into base and significant layers, in which the base layers contained the main content and structure information, and the salient layers contained the local area with relatively concentrated energy. The ratio of low-pass pyramid was also adopted to decompose the base layer into low-frequency and high-frequency layers. The corresponding fusion rules were designed according to the characteristics of the different layers. A sparse representation was used to express the relatively dispersed energy of the low-frequency base, and the rules of the maximum L1 norm and maximum sparse coefficient were weighted averages to retain different significant features. The absolute value of the high-frequency part of the base layer was used to enhance the contrast. Local variance was used for the salient layer to measure significance, and the weighted average was used to highlight the target area with enhanced contrast. Experimental results on the TNO datasets show that the proposed method performed well in both qualitative and quantitative evaluations. The method based on low-rank decomposition can enhance the contrast of the targets and retain rich details in infrared and visible fusion images.