Graph-based Salient Object Detection Via Multiple Priors

Salient object detection is an automatic process to extract the object region that captures human perceptual attention from input images. It simulates the information processing procedure of human visual system, and lays solid foundation for the further research on high level problems. This paper deals with the bottom-up, data-driven visual saliency detection without human’s subject intention, and designs detection algorithms for both single image and image set. Main works and innovations can be summarized into the following three parts:1. Saliency Model based on Boundary Prior with Foci of Attention: Based on the boundary prior, this model exploits the geodesic distance from each pixel to the boundary to measure the saliency through a graph structure. Meanwhile, a way of determining foci of attention based on maximal deviation from norm(MDN) is proposed to enhance the quality of saliency map. Experimental results on benchmark databases demonstrate the better performance of our proposed approach compared with several state-of-art methods.2. Saliency Model based on Multiple Priors and Energy Map: This model starts from both the salient object(foreground) and the background perspective, and takes advantages of respective priors to de?ne the corresponding saliency measurement. To better highlight the salient object, we adopt priors including contrast prior,center prior and grouping prior to measure the dissimilarity between different image elements. To better suppress the background, we exploit the boundary prior again and measure the pixel-wise saliency by the minimum seam cost where the seam is an optimal 8-connected path from the pixel to some boundary pixel. Experimental results show that our approach could obtain reliable saliency maps of high quality.3. Co-saliency Model for Image Sets: The main contribution of this model is to make existing saliency models work well in co-saliency scenarios. Given single image saliency maps, a two-stage guided detection pipeline led by queries is proposed to obtain the guided saliency maps of the image set through a ranking scheme. Experimental results on two benchmark databases demonstrate that the proposed framework well highlights the co-salient object while keeping the background suppressed. It can also deal with the case when irrelevant images are involved. Compared with existing models, the performance is enhanced greatly in terms of accuracy and e?ciency.4. Saliency Detection Applied in the Bokeh Image: An application called LeeBokeh is developed based on android for mobile devices. This application aims at making bokeh images accessible to common people, which greatly reduces the requirements on shooting devices, techniques and post-processing for generating a bokeh image. It automatically obtains the bokeh effect for an image.