Infrared and Visible Images Fusion with Multi-visual Cues

Abstract

Image fusion is an important technique which combines the original information from multiple input images into a single composite image. The fused images will be more beneficial to human visual perception or further computer processing tasks than any individual input. Most of the traditional infrared and visible fusion approaches perform the fusion on the assumption that the original information is measured by local saliency features such as contrast or gradient. There is little consideration of the “interesting” or “useful” information in global. In this paper, an infrared and visible image fusion method is proposed by considering the final aim of image fusion, the human visual perception and further image processing tasks. The fusion is implemented under the non-subsampled contourlet transform based image fusion framework. The low frequency sub-band coefficients which represent the intensity of the scene are fused with the weight map which is constructed by considering both visual saliency uniqueness and task-oriented objectness, and refined by spatial consistency with guide filter. The new fusion strategy ensures that the objects being “interesting” or “useful” are preserved in the fused image. Sixteen pairs of infrared and visual images are used to test the validation of the proposed method. The experimental results show obvious improvement of the proposed method in terms of both objective and subjective quality measurements corresponding to other methods.