An improved ROI searching system using dynamically selected color feature and object-based motion feature

In this paper, we propose an improved ROI searching system which selects color feature dynamically and extracts object-based motion feature based on SIFT algorithm. Previous systems have some problems that they cannot search some colors even if there are strong attentions in those colors. It is because previous systems only use one fixed color feature when using color-opponency. To overcome this problem, our system dynamically selects one that has strongest response among two opponent colors. Also we can draw moving rectangles of moving objects, and this leads our system to search ROI by object unit. In order to evaluate the performance of the system, we experimented with various images of still and video image, and verified its performance.