Journal of Xi'an Jiaotong University

Region Segmentation Based Multisensor Image Fusion Approach
Li Min^1,2,Cai Wei²,Tan Zheng¹
(1.School of Electronics and Information Engineering, Xi'an Jiaotong University, Xi'an 710049, China;
2.Department of Computer Science,The Second Artillery Engineering College, Xi'an 710025, China)

Abstract: Focusing on the most traditional image fusion algorithms that split relationship among pixels, a region based image fusion scheme was proposed. On the basis of image registration, source images were segmented into several regions, then two features of salience and visibility were extracted from each region to adaptively determine the fusion weight to get the fused image. The proposed method solves some of the well-known problems in pixel-level fusion, such as blurring effects and high sensitivity to noise, particularly when there is mis-registration of the source images. Experimental results show that it outperforms the discrete wavelet transform approach with 33% to 36% reduction in root mean square error, 0.9% to 53% improvement in standard deviation of gray level, 39% to 42% enhancement in mutual information, and 0.3% to 38% improvement in space frequency. The objective evaluation criteria coincide with the visual effect very well.
Keywords: image fusion; region segmentation; region feature; objective evaluation criteria