A Study on Automatic Filter Construction for Feature Extraction
Yano Yoshikazu, Tomonori Hashiyama, Shigeru Okuma · IEEJ Transactions on Electronics Information and Systems · 2003
A new automatic filter construction system for tracking an object is shown in this paper. Firstly, this system divides an input image with color features using mean shift clustering. And then, a new set of clusters are recombined into an optimal filter set which can extract inherent features. Furthermore, this system extract target using color features without the features of its shape and position. So even if the target object disappears in some frames, this system can detect it in the next some frames when the target appears again.