Research On Terahertz/Visible Dual-band Image Registration And Color Fusion

Terahertz imaging has attracted much attention in recent years and shows remarkablepotential in the field of security check. Terahertz has good ability of transmission, whichcan be used to detect dangerous objects concealed in the clothes; sorts of materials ofinterest for security applications such as drugs and explosives have characteristic terahertzspectra.Additionally, it poses no harm to the health of inspectors as passive terahertzimaging system doesn’t emit radiation actively.Information obtained by single sensor is limited which can not meet the actualdemand.Different sensors have different imaging mechnism and time,environment andangle are different which lead to image rotation,translation,scaling,etc.Image registration isnecessary before image fusion.Multiband image registration is more difficult comparedwith single band,especially for those images whose bands are far apart.Images of passiveterahertz obtained from actual only contain body contour and hidden items ofinformation.In this paper, we study the registration and fusion technology of terahertz andvisible image, in order to achieve the goal that the fused image can not only reflect checkappearance and security scenario,but also the information of hidden items.The research canmake up the shortage of the single band image and enhance the practicability of terahertzpassive security system.The main work of this dissertation include:(1) According to the existing registration method and with the characteristics of theterahertz images conbined, we select the image registration method based on gray level.This method require that registration of several band images have strong correlation in thegray level, therefore, using morphological processing method of optical image processingfor the visible light image and grayscale image can be obtained. This can be used for thefollowing image registration.(2) We propose a hybrid algorithm combining improved Powell algorithm with geneticalgorithm for terahertz and visible dual-band image registration. We use mutual informationas similarity measure to overcome the difficulty of extracting feature points from passive terahertz images. The approximate solutions of seven affine transformation parameters areobtained by genetic algorithm and they are set as the initial point of improved Powellalgorithm, and thenlocal search capabilities of improved Powell algorithm are used toobtain the exact solutions of seven transformation parameters.Three groups of experient aremade.Root-mean-square error is used as index to evaluate the registration results.Theexperimental results show that compared with genetic algorithm and improved Powellalgorithm，our algorithm can achieve higher matching accuracy. The study of this paper canimprove the accuracy of image registration.(3) We use image segmentation and image matching technology to make the hiddenitems of terahertz images which have been registrated more clear and the target is calibrated.This study puts forward a kind of passive terahertz image and visible image fusionalgorithm. In the algorithm, registrated terahertz image is segmented and pseudo-colorencoded. Taking the characteristics of terahertz image and visible image into account,principal component analysis is used to fuse the dual-band images and then IHS method isused to fuse visible image and the fused image above. The experiment results show that thisalgorithm can effectively fuse passive terahertz and visible image.