Stereo Saliency Analysis Based on Disparity Influence and Spatial Dissimilarity

Lijuan Duan, Fangfang Liang, Wei Ma, Shuo Qiu

Abstract: This paper presents a simple approach for detecting salient regions in stereo images. The approach computes saliency by considering three factors: disparity influence, central bias and spatial dissimilarity. Firstly, an image is split into equal-sized patches to be down-sampled. Next, disparity influence is estimated based on the disparity map. Besides, central bias value is assigned to every patch and spatial dissimilarity is measured between patches in reduced dimensional space. Thereafter, the product of all factors extracted from the image is computed. Finally, through a process of normalization, the saliency map is obtained. In the experiments four state-of-the-art methods are used for comparison with PSU stereo saliency benchmark dataset (SSB). The experimental results show that our method has better performance than the others for stereo salient region detection.