DCT 계수를 이용한 관심 영역 검출

Abstract

In this paper, a new method for extracting regions of interest (ROIs) using DCT coefficients is proposed. ROIs are detected by calculating the Bayes spectral entropy using DCT coefficients. In order to effectively extract ROIs, three candidate feature maps including ROIs are first generated from three images separated into Y, Cb, and Cr channels. Then, by combining three candidate maps, ROIs are finally extracted. Experimental results have shown that the proposed method achieves good performance of ROI extraction under various environments with multiple objects and cluttered backgrounds in natural images.