Study On Improved SIFT Image Matching Algorithm

Image registration is an important processing technique in image processing.It usually refers to the process of matching or superimposing two or more images acquired at different times,different sensors or under different conditions.It has been widely used in computer visual,remote sensing image processing and other fields.Image registration technology is mainly divided into two kinds of technology based on region and feature-based,in which the latter is better than the former image registration technology,the current mainstream image registration technology is also based on the characteristics.The main research contents of feature-based image registration technology include image feature extraction and feature matching.The research focus of the current algorithm and the improvement goal are mainlysignificance based on these two directions,so the research on image feature is of great.The SIFT algorithm proposed by Lowe in 1999 is an algorithm which is better in robustness in image feature extraction algorithm.It is also a relatively successful algorithm.This feature extraction algorithm can perform a good translation,and the illumination,scale Changes remain unchanged.Because SIFT algorithm has many excellent features,it has become a research hot spots in image registration field.This paper focuses on the shortcomings of SIFT algorithm,analyzes and proposes an improved method.This paper first introduces the theoretical knowledge of image feature extraction and image registration.Then,some corner extraction algorithms for image feature extraction are compared and analyzed.A new dimension reduction model is proposed to reduce the time complexity of SIFT algorithm.After the analysis and verification of different experimental data,the improved SIFT algorithm has good robustness,and the efficiency of SIFT algorithm is improved.