A Method of SIFT Simplifying and Matching Algorithm Improvement

Scale-invariant feature transform (SIFT) is a popular pattern recognition method in 2D-image because it can abstracts the features which are invariant to rotation, scale zooming, brightness changing. So it demonstrates a certain stability to objects subjected to view point changing and noise distribution. However, the dimension of the SIFT descriptors is too high, and its runtime is too long. Aiming at this disadvantage, this paper propose a new method to generate feature descriptor based on hierarchical region and treat different regions differently. Improved SIFT Algorithm reclassificates descriptor generating regions, using a circular area divide into 2×2+1 sub-regions instead of rectangular area in original algorithm. In the feature matching stage, setting different thresholds to 2×2 fan-shaped regions and 1 annular region, to achieve retaining right matched points as much as possible while removing wrongs. Comparing with the SIFT algorithm in some aspects, experiment results show that in the condition of fuzzy, light, rotation and affine transformation, improved SIFT algorithm can accomplish image matching test well and matching speed significantly improved.

I. Introduction

Computer vision is a simulation of biological vision, and positioning and image matching is an important part in computer vision applications. Matching algorithm based on features is an important branch in image matching, which has advantages of small amount of calculation and fast calculation speed. It is the most studied and widely used matching method currently.