Visual saliency detection by DCT coefficient dissimilarity

Visual saliency detection has recently become a highly active topic in image processing, due to its wide range of applications. This paper proposes a novel saliency detection model based on discrete cosine transform (DCT). The dissimilarity between image patches is evaluated using DCT low-frequency coefficients and is inversely weighted by the spatial distance between patches. An additional weighting mechanism is deployed that reflects the bias of human fixations towards the image center. The proposed visual saliency prediction model has been extensively evaluated on three image eye-tracking datasets and one video eye-tracking dataset. The experimental results demonstrate that the proposed saliency detection model outperforms the state-of-the-art models.