Accelerated Keypoint Extraction

Keypoints have become a fundamental feature in image and video processing. This article presents a wavelet-inspired structure to speed up the keypoint extraction process. The method restricts the extraction process to high variation areas solely, deemed likely to provide keypoints. The variation is calculated at different scales according to a hierarchical structure inspired from the Haar wavelets. This structure is also used for a fast access to the regions retained for processing. Experimental results are provided to show the efficiency of the method and the influence of the main parameters.