Study On Image Fusion Theory And Method

Image fusion is an important topic of image processing, which refers to synthesis and extraction of two or more multi-source image information, obtains more precise, comprehensive and reliable image of the same scene or target. Compared with the single sensor image, fused image maximizes the use of information on the various information sources, and improves the image resolution, enhances the sensitivity of the target image perception, role distance, measuring accuracy, anti-jamming capability, and is more suitable for human eye and computer-processing. Image fusion is usually developed on three different levels: pixel level, feature level and decision level. After study and analysis the domestic and international image fusion methods, the main concern of this dissertation is to study on pixel level fusion, feature level fusion and image fusion application. The main research is summarized as follows:(1) In order to directly fuse the image with different resolutions, a new different resolution image fusion method based on least mean square error is proposed, which is not strict with the resolution of source image, and enhances the robustness of image fusion technology.(2) Usually, while extracting the region in the region based image fusion method, the problem that size and number of the region are inconsistent may occur, introducing the joint region segmentation which is a good solution to the problem. At the same time, PCNN is introduced as the region similarity assessment, a feature level image fusion method based on region segmentation and PCNN is presented. The objective and subjective evaluation criteria are introduced in all the experiments of image fusion. Based on which the performance of image fusion is evaluated in various aspects including size of region, matching measure and the number of cluster. Experimental results show the proposed method improves the performance of fused image and the fused image is more consistent with human visual perception.(3) Image fusion is applied to the multimodal human face recognition utilizing the advantages of the thermal infrared and visual face image respectively, a new method using feature level image fusion and entropy component analysis is proposed for multimodal human face recognition. As details and character are increased, the fused images lead to better performance in the face recognition. The evaluation results indicate that the proposed method is efficient for the multimodal face recognition, even in the poor illumination and improve the recognition rate.