Fusion of infrared and visual images through multiscale hybrid unidirectional total variation
Yi Wang,Zhonghua Luo,Zhihai Xu,Huajun Feng,Qi Li,Yueting Chen
DOI: https://doi.org/10.1109/SIPROCESS.2016.7888220
2016-01-01
Abstract:As an important research area in image analysis and computer vision, fusion of infrared and visible images aims at delivering an effective combination of image information from different sensors. Since the final fused image is the demonstration of fusion process, it should reveal both source images' vital information distinctly. To achieve this purpose, an image fusion method based on multiscale hybrid unidirectional total variation (MHUTV) and visual weight map (VWM) is proposed in this paper. The MHUTV combines the feature of extracting the details from images and the capacity of suppressing stripe noise, which leads to a more ideal visual effect. The MHUTV is a multiscale, unidirectional and self-adaption image decomposition method, which is used to fuse infrared and visible images in this paper. The visual weight map aims to reveal attention drawing distribution of human observer. It provides a subband fusion criterion, which can guarantee the highlighting of interesting area from infrared and visible images. Firstly, multiscale hybrid unidirectional total variation is discussed and used to decompose the source images into approximation subbands and detail subbands. Secondly, the approximation and details subbands are respectively fused by a fusion rule based on visual weight map. Finally, the fused subbands are combined into one image by implementing inverse MHUTV. The results of comparison experiments on different sets of images demonstrate the effectiveness of the proposed method.