A spectral algorithm based on weighted sparse coding for visual saliency detection

A spectral algorithm using weighted sparse coding is proposed for visual saliency detection in this paper. This algorithm is able to improve the accuracy of the traditional spectral saliency detection approaches while preserving their advantage of fast processing speed. Based on the traditional sparse coding algorithm, the sub-codes are weighted according to their incremental coding length. Then, the image is encoded by weighted sparse coding instead of directly transforming the raw images into frequency domain. Finally, we improve the multi-channel method developed in a recently published algorithm called image signature through information theory. Our method yields the saliency map with the form of the Shannon self-information. The experimental comparisons between the proposed and 9 state-of-the-art approaches and the analysis of complexity of our algorithm demonstrate the effectiveness and efficiency of our method.