Based On The Passive Millimeter Wave Imaging Methods Of Image Registration And Fusion Research

Passive millmeterwave (PMMW) imaging technology utilizes radiant energy differences in millmeterwave band to obtain images of scene. It can work in all-weather condition, detect concealed weapons under cloths and cope with military disguise. However, millmeterwave images are always low in resolution and inherently blurred, what’s more, less information, which deficiently and coarsely describe the ture scenes, would be obtained in simplex sensor model. Image fusion technology can synthesize advantages of optical sensor such as high accuracy and intuitive vision, thus achieve mutual benefits of multi-sensors and improve image quality effectively.It has become a frontier and hot issue research technology.The key research of image fusion is focused on overcoming image misalignment, information redundancy and image noise. Considering that and directing at fusion work for PMMW images and optical images, this article studies a silhouette-featured registration method based on Fourier-Mellin transform (FMT) and an image fusion method based on expectation maximization (EM) estimation theory, mainly as follows:1. To deal with gray-level inconsistencies between images, we synthetically utilize several basic image processing procedures to extract silhouette features as registration factors, which can effectively guide FMT based registration method.2. When applying FMT based method, Inaccuracy occurred in roation estimation, and false peaks of cross-power spectrum (CPS) disturbed translation estimation. To solve that we present a frequency domain enhanced method and a multi-peaks searching strategy, which can optimize estimation of parameters and calculation accuracy, compared with traditional FMT based registration method.3. Analyzing image characteristics such as local reversal and local offset between gray-level features of images, we novelly establish a sensor imaging model and take image noises into consideration simultaneously.This model can simplify sensor mapping relationships between images and underlie the implementation of estimation based fusion method.4. We study an estimating methods using Expectation Maximization algorithm (EM) to calculate fuse parameters, then deduce and verify the whole algorithm. To avoid large computation and fit model better, window-based and region-based data preprocessing methods are adopted. Both of them can achieve good fusion results. Moreover, the region based method performs better in subjective and objective fusion quality evaluation, and owns better anti-noise capability.We simulate on groups of images to demonstrate effectiveness of methods in article. Results show that the improved silhouette-based FMT registration method can register image more effectively and presicely, and region-based EM fusion method can reach better image quality and anti-noise property than window-based fusion method and wavelet fusion method.