A Simple and Fast Feature Extraction and Block- Matching Based Feature Searching Technique

Feature extraction and tracking is the first step in variety of computer vision applications like stabilization, stitching, optical flow etc. For online applications such as stabilization of online video stream, we need fast and simple techniques for detecting and searching features. In this work, we propose a very simple feature extraction and detection technique, that is a mixture of block-based and feature based techniques. We treat some selected blocks as features and apply fast block matching algorithms and then again treat the positions of blocks as features to apply the motion model fitting. We achieve 50%, average improvement over naive block-matching based feature searching techniques.