CORNER-BASED KEYPOINTS FOR SCALE-INVARIANT DETECTION OF PARTIALLY VISIBLE OBJECTS

摘要

Local features (also known as interest points, keypoints, etc.) are a popular and powerful tool for matching images and detecting partially occluded objects. While the problems of photometric distortions of images and rotational invariance of the features have satisfactory solutions, satisfactorily simple scale-invariant algorithms do not exist yet. Generally, either computationally complex methods of scale-space (multi-scale approach) are used, or the correct scale is estimated using additional mechanisms. The paper proposes a new category of keypoints that can be used to develop a simple scale-invariant method for detecting known objects in analyzed images. Keypoints are defined as locations at which selected moment-based parameters are consistent over a wide range of different-size circular patches around the keypoint, While the database of known objects (i.e., the keypoints and their descriptions) is still built using a multi-scale approach, analyzed images are scanned using only a single-scale window and its sub-window. The paper focuses on the keypoint building and keypoint matching principles,. Higher-level issues of hypotheses building and verification (regarding the presence of objects in analyzed images) are only briefly discussed.