Detection of visual saliency in Discrete Cosine Transform domain

In this paper, a novel visual saliency model for detecting visual saliency from compressed images is proposed. To this end, we use a measure of camera focus on DCT (Discrete Cosine Transform) frequency domain. In order to effectively detect visual saliency, the proposed method calculates Bayes spectral entropy of DCT coefficients. Consequently, three focus maps are calculated from Y, Cb, and Cr channel images. By combining these focus maps, a saliency map is constructed. In experimental results, the proposed method shows better performance of visual saliency detection, compared with conventional methods.