|
Image matching is aligning two or more images of the same scene in the space. It is a important problem in the research of computer vision, it is a key step from image processing to image analysis. at the same time, with the application lucubrated,it is more and more exigent to the request of adaptability, precision and the real time of Image matching. To make a target of the request, this dissertation presents a FAST image matching algorithm based on feature points. The main work of this paper is as follows:Firstly, we compare the capability of the classical feature points detection algorithm, such as Harris, SUSAN, FAST. It is known that the effect of the FAST algorithm is better than the others, because the FAST algorithm doesn’t have scale invariant characteristic, and it is very hypersensitive to noise. We can find a method which can get over its failing with improving the FAST algorithm. A detailed performance comparison of different scale invariant feature point detectors is given.Secondly, introduce the flow of SIFT algorithm, taking into the SIFT algorithm in real-time is bad. Making that the correspondence of feature points is the dot metrix of feature points, this thesis presents a FAST image matching algorithm based on Zoser-FAST feature points.Finally, when change happened in picture, such as scale transformation, noise, dark, compress, brightness, we compare the algorithm of this thesis with SIFT and SURF, The results show that processing speed of new algorithm is better than the other two. Besides,we analyse the capability of new algorithm when transformation happened in picture, so that affirm the applicable area of new algorithm.
|