Research on image detection and matching based on SIFT features

Scale-invariant feature transform (SIFT) is a kind of computer vision algorithm used to detect and describe Local characteristics in images. It finds extreme points in scale-space and gets its coordinate, scale, orientation, which in final come into being a descriptor. This paper studied the theory of SIFT matching, use Euclid distance as similarity measurement of key points and use RANSAC to eliminate mismatches. The result shows that SIFT algorithm is invariant on rotations, translations and scaling and SIFT features have strong matching robustness for radiation transformation, perspective changes, illumination changes and noises. This paper also compare different results obtained by different ratio threshold and finally set 0.6 as the best value considering the balance number of matched points and matching accuracy. It is important to image recognition application.