Reserch And Implementation Of Multi-source Image Fusion Algorithm For Passive Millimeter Wave Imaging System

The Passive Millimeter Wave(PMMW) imaging system achieves imaging through detecting the difference of natural radiation energy between the target and the experimental scene. The PMMW imaging system has the penetrability, so it can detect hidden dangerous targets, and the system will not radiate the electromagnetic wave in the process of imaging, which means that it is harmless to human body. Therefore, Passive Millimeter Wave imaging technology is widely used in the fields of surveillance and security. The image obtained by the Passive Millimeter Wave imaging system contains less high-frequency information, so it will affect the image’s visual determination. Through combining the millimeter wave image and optical image, the fused image can both reserved target information hidden in the millimeter wave image, but also has a complex optical image background.The research of this paper relies on actual research projects. Firstly, the framework of image fusion is studied. And then, the pre-processing algorithm of image fusion is analyzed, including de-noising and image segmentation. At last, according to specific application, the fusion method of the Millimeter Wave image and optical image is determined. Research contents include:1. Analysis of the basic theory of passive millimeter wave imaging detection technology, including the characteristics of the millimeter wave radiation and radiation model. Some typical image fusion methods were summarized, the evaluation criterions of image fusion quality are given here.2. The Millimeter Wave image preprocessing algorithm is studied. On the basis of the median filtering, an algorithm which can adaptively detect noise points is studied, according to the results of detection, removing the noise in the image.3. To extract the interest target area in the millimeter wave image, commonly used image segmentation method is studied. And then, mainly introduced the image segmentation algorithm based on Gaussian Mixture Model. The traditional maximum entropy image segmentation algorithm has the problem of low segmentation threshold, aiming at this problem, the process of pre-segmentation is used to improve the segmentation accuracy based on the maximum entropy algorithm, in the pre-segmentation, some points with low gray value is removed, experiment results show its effectiveness and reliability.4. An image fusion algorithm based on nonsubsampled contourlet transform is studied, the algorithm treat the result of image segmentation as the criteria to determine whether the image pixel belongs to the target area of interest, according to the segmentation result, different fusion strategy are applied for ROI and background region. At last, the sub images of different scales are reconstructed by NSCT, the fusion image is obtained.