Two-layer average-to-peak ratio based saliency detection

Visual saliency plays an important role in pattern recognition tasks such as rapidly seeking prominent regions in a complex scene to discover the meaningful objects. In this paper, we present a new method to detect visual saliency from an image. This saliency is modeled as two parts, i.e., average-to-peak ratio (APR) saliency and chrominance-aware (CA) saliency. The first term is designed to describe the global contrast, which is computed based on pixel-level saliency maps. To compute the CA saliency, the luminance component is first removed by subtracting it from each color channel. Then the difference follows the APR saliency computation. Finally, a two-layer saliency model is built by combining the two saliency maps. To evaluate our proposed method, we do extensive experiments on three well-known image data sets including MSRA image set, PASCAL VOC image set, and human fixation dataset. Experimental results show that our method outperforms the state-of-the-art methods and achieves the good performance on the visual saliency detection task.

► We present a two-layer average-to-peak ratio based saliency detection method. ► Average-to-peak ratio saliency and chrominance-aware saliency are defined. ► The APR saliency is designed to describe the pixel-level global contrast. ► The CA saliency extracts the chromatic saliency by removing the luminance component. ► Our model achieves the good performance on the visual saliency detection task.