Chinese Journal of Computers   Full Text
Title	Multisensor Image Adaptive Fusion Based on Nonsubsampled Contourlet
Authors	CHANG Xia1) JIAO Li-Cheng1) JIA Jian-Hua1),2)
Address	1)(Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education of China, Institute of Intelligent Information Processing, Xidian University,　Xi’an 710071) 2)(School of Information Engineering, Jingdezhen Ceramic Institute, Jingdezhen, Jiangxi 333002)
Year	2009
Issue	No.11(2229—2238)
Abstract &   Background	Abstract An adaptive fusion method of multisensor images based on nonsubsampled contourlet transform is proposed in this paper, which can select the fusion weights of the low-frequency coefficients adaptively via golden section algorithm. The nonsubsampled contourlet transform is a flexible multi-scale, multi-direction and shift-invariant image decomposition, which is suitable for representing images bearing abundant detail and directional information. This is employed for fusing the directional high-frequency coefficients. For the directional high-frequency coefficients, the higher adding level of the directional subbands is used to select the better coefficient for fusion. The nonsubsampled contourlet transform can also avoids introducing ringing artifacts to fused images compared to ordinary method. Experimental results show that the proposed method achieves better fusion efficiency compared to image fusion methods based on Laplacian pyramid transform, wavelet transform, stationary wavelet transform and contourlet transform respectively. Keywords image fusion; adaptive; golden section; nonsumpled contourlet transform; multiscale geometric analysis Background The work is supported by the National Natural Science Foundation of china (60702062), the National Basic Research Program (973 Program) of China (2006CB705707), the Provincial Natural Science Foundation of Shaanxi of China (2007F09), the National Research Foundation for the Doctoral Program of Higher Education of China (200807010003) and the Program for Cheung Kong Scholars and Innovative Research Team in University (IRT0645). This group has been working on image processing, natural computing and machine learning for many years. So far, parts of the research results have been published in important academic journals, among which about 100 papers are indexed by SCI and EI. Multisensor image fusion has important applications in remote sensing, computer vision, medical image analysis, and so on. In recent years, plenty of successful fusion methods based on transform have been put forward to merge multisensor images. These methods employed multiscale transform as the tool for extracting images salient features, and that the characteristics of the transform tool affect the quality of the fused images evidently. The existing multisensor images fusion methods usually adopt weighted sum operation to set low-frequency subbands fusion weights. However, the relation between the objects merged in multisensor images isn’t simply linear weighted. And weighted sum operation needs obtaining the weight of each object beforehand, which leads to great subjective preference. So an optimal method to select the low-frequency subbands fusion weights adaptively is needed. An effective image presentation has the requests of bearing shift-invariance and apperceiving the geometric property of nature scenes effectively. The new multiscale geometric analysis tool, nonsubsampled contourlet transform, is applied to multisensor images fusion domain in this paper. The golden section algorithm is used to select the optimal fusion weights of the low-frequency coefficients adaptively. The high-frequency directional subbands of nonsubsampled contourlet transform capture the salient features of multisensors. The higher adding level of the directional subbands is used to select the better coefficient for fusion in this paper. Experimental results show the proposed method achieves better fusion efficiency.