Research Of Automatic Image Registration Based On SIFT And Corner Detector

The existing image registration methods can be divided into two parts. One is registration method based on gray level. Another is registration method based on feature. The latter, which is now widely used, can accelerate registration efficiency by local match. SIFT algorithm is a feature point matching algorithm belongs to registration method based on feature. In my paper, I propose a improved SIFT algorithm in order to reduce running time and improve matching accuracy.In recent years, image registration algorithm based on SIFT algorithm has become one of the classic image registration algorithms. This algorithm is used to match images with scaling change, rotation change, illumination change and viewpoint change, which can get very high accuracy. However, more characteristics of multi-points we extract, higher dimensional vector we describe, that will result in more running time and lower efficiency. This paper proposes a feature extraction algorithm combine with SIFT operator and corner detection operator. We think that if gradient directions are in opposite directions, gradient magnitudes can be subtraction. So in that way vector dimensions cut in half. Finally, speed of image registration accelerates.Steps of Image registration based on Feature Points are feature extraction, formation characteristic descriptor, finding feature matching points and image registration. In this paper, we intensively focus on feature extraction and formation characteristic descriptor. We also develop methods for feature point extraction and methods for descriptor generating, which are both based on SIFT algorithm.In this paper, under Microsoft Visual Studio 2008 Development Platform, we choose OpenCV library to develop and implement the above methods and to construct a image registration system. Experiment result concludes that our improving algorithm proposal can develop running speed and registration rate for sure.