Image Feature Point Matching Based on Improved SIFT Algorithm

Abstract

An improved SIFT algorithm (SIFT-BRISK) is proposed for the acquisition of image feature points due to the multi-scale, noise, light intensity and rotation between images, which cause image matching is not ideal and feature point matching time is too long. The feature points are collected in the form of DOG pyramid. By comparing the pixel points of the upper and lower layers of the DOG pyramid at multiple scales, we find the extreme points as our feature points. When constructing the feature point descriptors, we abandon the traditional basis. then use the BRISK algorithm to construct concentric circles of different radii in uniform sampling mode as sample points to generate binary descriptors, this way and traditional SIFT descriptor generation Compared to the faster speed, the sample points are more unique because of the different levels of filtering. In the image matching part, the traditional violent matching is matched and innovative. From finding the shortest Hamming distance between the different pictures, the group matching method is used to find the Hamming distance with the nearest set of matching descriptors. The description sub-pair of the shortest Hamming distance greatly reduces the time for matching between binary descriptors and improves efficiency.