Visible and Infrared Image Fusion via Superpixel Segmentation and Salient Region Detection

To get better performance of fusion, accurate segmentation and suitable clarity measures always play a significant part in region-based image fusion. In this paper, a novel region-based method is proposed for visible and infrared image fusion, which can effectively address the issues of inaccurate segmentation and unsuitable clarity measures. On the one hand, by applying a superpixel segmentation algorithm such as linear spectral clustering (LSC), more homogeneous regions can be acquired for fusion and the risk of generating artifacts is also decreased. On the other hand, a salient region detection method utilizing the cues of compactness and local contrast is carried out to get the saliency information of each superpixel. Moreover, by comparing the saliency values of the superpixels at the same position in visible and infrared images, the superpixels with larger saliency values are selected region by region and then the final fusion image can be generated. The experiments are conducted on six sets of visible and infrared images. In addition, four other mainstream fusion approaches are adopted for comparing the performance of fusion. Experimental results demonstrate that our proposed method is effective in visible and infrared image fusion problems.