An Improved ORB Algorithm of Extracting Features Based on Local Region Adaptive Threshold

Abstract

For the features extraction of image, an improved Oriented FAST and Rotated BRIEF (ORB) algorithm of extracting features based on local region adaptive threshold is proposed, which not only can it solve the problem that the traditional ORB feature extraction algorithm cannot adapt to the local brightness change of the image, but also solve the phenomenon that the extracted feature points exist clusters. Firstly, extracting local region adaptive threshold features on each pyramid image based on constructed an image pyramid. Then, the feature points are divided by the quad-tree algorithm and the direction and the descriptor of the feature points are calculated. After that, the Fast Library for Approximate Nearest Neighbors (FLANN) is used to the match feature points, the mismatch points are eliminated by Lowe's algorithm and rotation consistency. Finally, using the Random sample consensus (RANSAC) to get the fine matching image. The method proposed in this paper is carried out on the Oxford images. Experiments show that the proposed method can extract more stable feature points under different fuzzy, illumination and compression conditions, which improve the matching accuracy of feature points.