Saliency Detection for High Dynamic Range Images via Global and Local Cues

Abstract

Aiming at the problem that saliency detection algorithms for low dynamic range (LDR) images are unsuitable for high dynamic range (HDR) images, we propose a new saliency detection method for HDR images, where the global and local cues are considered. Firstly, according to human visual perception of high dynamic range content, the luminance and chrominance are processed respectively. Secondly, the bottom-up saliency map (BU-SM) is obtained by the global information. Then, we construct the foreground and the background codebooks based on the BU-SM, and use the sparse coding to get the top-down saliency map (TD-SM). Finally, in order to well account for the global and local factors, BU-SM and TD-SM are combined to get the final saliency map of HDR images. The experimental results show that the proposed method is superior to the state-of-the-art methods.