Using a Hierarchical Keypoint Model for Object Recognition

Abstract

It is important to detect keypoints and calculate their descriptions before engaging in local keypoints matching between a pair of images for object identification. Accurately describing landmarks is crucial for many vision-based applications, including 3D reconstruction and camera calibration, structure from motion, picture stitching, image retrieval, and stereo pictures. To address this issue, this paper presents (1) UFAHB, a robust keypoints descriptor using a cascade of Upright FAST -Harris Filter and Binary Robust Independent Elementary Feature descriptor, and (2) a comprehensive performance evaluation of UFAHB descriptor and other state-of-the-art descriptors using a dataset extracted from images captured under different photometric and geometric transformations (scale change, image rotation, and illumination variation). The gathered experimental results show that the integration of the UFAH and BRIEF descriptors is fast in execution time and robust against variations in illumination.