Research On Algorithm And Application Of Feature-based Image Registration

The image registration technology is a basic task of the computer vision and have widerange of applications. The image registration technology can be classified into two classes:pixel-based registration and feature-based registration, each with its own characteristics.Pixel-based registration’s anti-noise performance is poor. With the proposed of SIFTalgorithm the Feature-based registration gradually become a hot topic. This algorithm isused to match images with scaling change, rotation change, illumination change andviewpoint change, which can get very high accuracy. However, more characteristics ofmulti-points we extract, higher dimensional vector we describe, that will result in morerunning time and lower efficiency. The process of the feature-based registration can bedivided into three sub processes: key-points detection, descriptor construct and descriptormatching. In order to solve those problems, we improve the key processes.This paper proposes a key-points detection algorithm with SIFT operator and cornerdetection operator. We think that if gradient directions are in opposite directions, gradientmagnitudes can be subtraction. So in that way descriptor dimensions cut in half. Finally,speed of image registration accelerates.At last research on the key technology of image registration application, construct achange detecting system based on image registration.