Subpixel Registration Of Infrared And Lll Image Fusion Technology Research

Image registration technique is an important research topic in the field of image processing, and has been widely used in the fields of remote sensing image, image stitching, medical image, military night vision etc. The infrared and low-light-level (LLL) image registration is the most pressing issue of multi-source image fusion. There is a great difference between the infrared and LLL images arising from quite different characteristics manifested in the same scene, which brings a great challenge to the infrared and low light level image registration.In this thesis, the infrared and LLL images in the image registration experiments are from IRFPA and super second generation ICCD detector. An optical system parallel to the optical axis is chosen to ensure the acquirement of image informations. The optical path from low-light detector to the target imaging is adjusted according to the principle of optical axis parallel of the optical wedge, so as to increase the scene coincidence degree of images obtained by the two different detectors.Because of the grayscale differences between infrared and LLL images, the corner detection method is selected to detect the image feature points. Moravec, Harris, and improved Harris corner detection algorithms are emphatically analyzed, and the improved Harris corner detection algorithm is chosen as the best algorithm after trial comparison. There are some displacements, rotations and scalings between the images of infrared and LLL detectors, which may even produce distortion, therefore, an affine transformation model with six parameters is chosen. The images are precisely matched by using the Hausdorff distance to verify the correctness of the matching points and exclude the incorrect match points. With the obtained matching points, the parameters of the affine transformation are fitted by the method of least square. Then, the matched LLL image is obtained by resampling the LLL image based on the infrared image.By use of super-resolution image reconstruction, the sub-pixel registration between the infrared and LLL images is carried out to achieve a registration accuracy of0.5pixels. The image reconstruction methode classification, New Edge-Directed Interpolation (NEDI) algorithm, and the improved NEDI algorithm are introduced in detail, and the improved fast NEDI algorithm is chosen for the optimal interpolation algorithm. The super-resolution reconstruction of infrared and LLL images is firstly achieved to obtain high-resolution images, then the image corner detection and registration are used to obtain registration of high-resolution images. Last the image down-sampling is implemented to achieve the registration accuracy of0.5pixels. The combination of image registration and super-resolution reconstruction can obtain better sub-pixel registration requirements of infrared and LLL images.