Reliable and Accurate Pattern Search by Combination of Absent Color Indexing with Correlation Filter
Ying Tian, Shun’ichi Kaneko, So Sasatani, Masaya Itoh, Ming Fang · 2019
Color-based image processing has been tried as a mean near to human vision, so they have been utilized to realize a likeness to human eyes, for instance in flexible extraction of prominent features. In this paper, a novel approach to image registration called Absent color indexing (ABC) is proposed to address the problem of robust pattern search. Color histogram-based methods have shown favorable performance in the conditions of object rotation, deformation, and occlusion. Most of them have had good performances, but their weakness to fail in distinguishing similar but different objects in the scene. The proposed approach works well in these situations by using absent colors that have relatively low-frequencies or nonexisting colors in color histogram bins. In order to obtain a more accurate positioning, we extended our approach to combine with a correlation filter. Experimental results on Mondrian random patterns and the real-world images show that the proposed ABC has rather high performance for distinguishing similar objects.