A block-based clustering technique for real time object detection on a static background

Block-based statistic methods for object detection are used as tools for fast object discrimination. However, their performance degrades when it comes to image alterations of a large scale. This paper presents a block-based clustering technique applied on blocks of the absolute difference. The proposed method presents the additional benefit of increased robustness against large-scale image alterations. Two variations of the algorithm are presented: one applied on grayscale image pairs and one exploiting chromatic variations in color images.