|
With the development of science, image matching has not only become a very important technology in the field of modern information procession but also been the basis of some image analyzing technology, and has been applied in many other fields.In various fields such as aircraft cruise guided, remote sensing, end-guidance of missile projection system, visual control, medical imaging, and computation of visual motion, there may be a large rotation between the template and the reference, so far there has not been a method using at large against arbitrary rotations. Thus, it is significance and valuable to research on image matching method against arbitrary rotations.There are many matching methods for rotated images, which, however, have some drawbacks, such as low speed, low reliability, inability for arbitrary rotations, and so on. On the basis of analyzing their merits and drawbacks, this thesis presents a new algorithm that could make up for the shortages of existing approaches. Its main contents are as follows:1. Several typical template-matching strategies are summarized. Various matching and registration methods for rotated images and their rationales are analyzed and their merits and drawbacks are compared.2. This thesis focuses on the study on an image-matching method that based on the translation invariance of Fourier transform, which transforms the rotated status to translation status by changing the image structure, and then calculates the Fourier transform magnitude spectrum of the changed image and correlates the magnitude spectral values instead of original gray levels. For speed consideration, hierarchy-searching strategy with two steps is applied throughout the matching.Many experiments have been carried out for the proposed method and analysis have beenmade on its performance of anti-rotation, matching speed, and matching correctness. The experimental, results show that the algorithm can against arbitrary rotation with a rather high speed and a rather high matching rate, and it is robust to scale transform with low noise.
|