Fusion Algorithm Research Of Passive Millimeter-wave Image And Visible Image Based On Statistics Theory

Passive millimeter-wave(PMMW) imaging technology obtains images through the difference of intensity in millimeter-wave band that targets and scenes radiate. It has many advantages, such as the great performance in penetrating of the textile and the smoke of battlefield, the huge difference of radiation energy between metal target and surrounding environment and no radiation damage to human body, which makes it has vast application prospect in security checking and battlefield reconnaissance. However, the low spatial resolution and Signal Noise Ratio(SNR) of PMMW image can’t describe the details of scene accurately, which cannot meet the requirement of application. Image fusion technology between PMMW and optical images can synthesize the detection ability of PMMW image and the superiority of optical image to break through the limit of single sensor. It has become a frontier and hot issue research technology.This thesis studies the preprocessing and segment of PMMW image, and the registration method based on Fourier-Mellin transform(FMT); On the other hand, this thesis gives an image fusion algorithm for PMMW and Optical image based on image model. The mainly contents are as follows:(1) Research the image enhancement method. Studied the target extraction and segmentation algorithm according to the feature of PMMW image to prepare for image fusion.(2) Studied the image registration method based on Fourier-Mellin transform(FMT). We give a solution with adding edge fuzzy window function for the problem of failed registration, which eliminates the phenomenon of false spectrum peak effectively. Meanwhile, we present an improved method of spectrum peak searching by fitting of parabola interpolation, the simulation experiment verified theirs correctness and effectiveness.(3) PMMW and optical images have huge difference in gray-level but in consistencies with image contour, so this thesis propose a registration method based on their silhouette features and Fourier-Mellin transform. The simulation shows its effectiveness.(4) Studied the image formation model, Gaussian mixture model and Expectation-Maximization(EM) algorithm, Given the image fusion model combined image formation model and Gaussian mixture model and derived it’s solution formula by EM, which provides important theories for the subsequent image fusion.(5) Aiming at the fusion of the PMMW and optical image, studied the classic image fusion method based on wavelet and beyond wavelet and given an image fusion algorithm based on region segmentation and image fusion model. The fused images using the proposed method have high contrast, clear texture, in line with human visual habits and suppress the noise subjectively; Meanwhile, The fused images also get a better score in objective evalua tion criterion.