Study On Registration-algorithm In High Dynamic Image Of Moving Objects

In recent years, as a multi-cross-cutting, multi-disciplinary field of the problem, high-dynamic range image registration has been growing attention to research it. And its branches have important, wide-ranging and significant meanings. High dynamic range image can effectively avoid the highlight and saturation issues caused by the long exposure time and the region of the high reflection coefficient, but also to maximize the effective use of multiple conventional low dynamic range images'information and data. It is greatly significant to take and use the moving objects in the environment.Different from conventional low dynamic range image of the object tracking and registration, in high-dynamic range images, the texture, color, characteristics of objects in the environment will change with the exposure time change, due to the multiple different exposure images. Generally the conventional match and registration algorithm does not apply on them.Through to study the change of moving objects in high dynamic range with exposure time and environmental changes, and to combine with high dynamic range image of their own characteristics, a serious of research and practice for high-dynamic range image registration algorithm is of great significance.Based on the study of moving objects in high dynamic range images with the changing of the exposure time and environmental, the limitations——texture features usually changes with the light and thus can not maintain consistency in the series of images, and drawbacks——the function curve of the camera should be calculated before the alignment algorithms, of high dynamic range image registration application for the conventional matching and registration algorithms are analysis in deep. And use them to draw contrast to the shortcomings caused by excessive reliance on various features result to sensitive to the multi-exposure image, of existing conventional algorithms and to identify the corresponding point of breakthrough——deal the image with the pixel binarization and frequency domain processing methods, combined with high-dynamic range image features , principles——using the camera response curve and characters, to research and improve of high-dynamic range image registration algorithms.In this thesis, we: Study a high-dynamic image registration algorithm (MTB algorithm) based on the rapid bit operation of computing binarization image. Since MTB algorithm for binarization threshold selection is very sensitive, binary images which pixel values are relatively concentrate are always lead to significant error caused by the introduction of the few noise. We improve a semi-adaptive threshold method for the algorithm to improve the MTB's median threshold classification based on the experiment, and reduce the effects for the algorithm due to the threshold selection.Study a Fourier—Mellin High Dynamic image detection algorithm based on the camera response function to restore the E value (brightness value). The data obtained by using the camera response function to restore the E value in a darker image, is often fluctuate due to which noise and signals are very close, through experiments, this paper has improved a simple and fast iterative search method to prevent introducing the unnecessary noise in a very dark areas, and the a standard and default function used to balance the too bright or dark pixels in the region has a very important role.Achieve results show that, in general, in condition of high-dynamic range planar images registration with no parallax, the MTB algorithms and Fourier—Mellin algorithm have good performance; in condition of high-dynamic range three-dimensional images registration with parallax, the Ghost region would lead from a variety of algorithms; MTB algorithm and the Fourier—Mellin algorithm have their own advantages in terms of complexity and robustness in the algorithm.