Research On Multiband Image Fusion Algorithm Based On Clustering And Multi Scale Decomposition

The Passive Millimeter Wave(PMMW) imaging system can detect consealed targets from human body based on energy difference radiated by objects. PMMW imaging system is widely used in airport security, monitoring of key areas and investigation of battlefield. Because of the limitation of antenna performance parameters, the MMW image is relatively fuzzy and resolution is poor. These shortcomings stand in the way of future research of image detection and image fusion. Based on multi-band image fusion technology, the MMW image and visual image of very same scenario are fused to integrate advantages of multi-sensor imaging. The fused image combines the complementary information and reduces redundant information of each single sensor image, which is convenient for human observation.This paper is based on specific research project. Clustering analysis technology is first studied to segment the MMW image to extract the region of interest(ROI); then, a multi-band image fusion method based on image segmentation and multi-scale decomposition is proposed. Research contents include:1. The PMMW imaging theory, model and multi-band image fusion theories are studied. The millimeter wave imaging principle and imaging features are analyzed. Some typical multi-band image fusion methods were summarized, the evaluation criterions of image fusion quality are given here.2. Clustering analysis technology is studied. The clustering model and the similarity measurements of objects are expounded. K-means clustering algorithm is introduced to segment MMW image; experiment results show its effectiveness and reliability.3. The fuzzy C-means clustering algorithm and clustering of Gaussian mixture model are studied. By optimizing the objective function, the introduction of neighborhood constraint information, the standard fuzzy C-means clustering algorithm is improved. Based on the idea of probability distribution, the histogram of image pixels is fitted. The gaussian mixture model and the expectation maximum(EM) algorithms are introduced for segmentation of MMW image. ROI is extracted and binarization image of target and background is obtained by clustering algorithm.4. A multi-band image fusion method based on image segmentation and multi-scale decomposition is researched and proposed. After MMW image is segmented by clustering algorithm, the MMW image and visible image are both transformed by nonsubsampled pyramid(NSP), different fusion strategy are applied for ROI or background region. At last, the sub images of different scales are reconstructed by NSP, the fusion image is obtained.Algorithms of this paper are simulated by Matlab and Visual Studio software under Windows operating system. Images are obtained by PMMW imaging system; the experiment results show the effectiveness and stability of the algorithm.