Research Of Image Fusion Method Base On Image Matting

Because of the diversity of sensor category and the limitation of sensor imaging, the images of different sensors have different characteristics. While obtaining some scene by one single sensor, the image obtained cannot contain all the detailed information in the scene. In order to solve this problem, we can use many different sensors to capture one scene and fuse the captured different scenes and pictures to make the fused picture contain all the information in the scene. At present, the fusion methods are mostly in allusion to specific scene fusion. These methods have some limitations. For example, the multi-focus scene fusion method only takes good effect in mutil-focus picture, but the fusion effect is not good while used in other scene. Therefore, this work proposes a fusion method based on the KNN-Matting according to the problems described above. The proposed method is based on the KNN-Matting. The main work in this paper is as following:First, the principles of image matting and KNN-Matting image matting method, and the matting based multi-focus image fusion are briefly reviewed.Second, on the foundation of image matting deficient usage in multi-focus fusion method, this work proposes a KNN matting based multi-mode fusion method, which can be used in the multi-focus fusion and others fusion applications. The main idea of this method is to treat the salient areas of each original image as the foreground areas, and then use the KNN-Matting method to extract out the foreground areas from the original images, Finally, the salient areas are combined together to obtain the final fusion results. Experimental results show that the proposed method works effectively in multi-focus, multi-exposure, and multi-modal image fusion.Finally, in order to further improve the fusion performance, an improved two-stage KNN matting based image fusion method is investigated in this work. The proposed two-stage fusion method can effectively solve the in selecting initial foreground pixels, which is an important step of the KNN matting. Experimental results show that the improved two-stage fusion method achieves better fusion performances against recently suggested fusion methods and original of KNN-Matting based fusion method.