Depth Enhanced Saliency Detection

Observing an image, humans unconciously judge the importance of image regions and focus attention on important parts. This kind of selection is based on the visual attetion mechanism. As the regions focused by most ones are the same within a period of time, searching the attetion regions can be a substitute when understand the human behavior pattern. The detection of the focus regions is called "saliency detection". Saliency detection is first modeled by imitating the reaction of human visual center system for visual features contrast. Nevertheless, saliency detection is first processing on the 2-dimensional for the short development time and the limitation of the tecnology.However, human visual system understand the world though 3-dimensional images. The past 2-dimensional saliency detection methods are all lack of the handling of the important depth information. Althought these studies contribute to the understanding of human behavior pattern, ignoring the depth information is still a simplified version of the detection problem. To complete the visual perception and search the effetion of depth information for the saliency detection, we propose a depth enhanced saliency detection method. This method has three cues. The first two cues are the reflection of the visual saliency stimulation theory on both color and depth space. The last cue "spacital bias" is a combination of depth and "center bais" which is widely used in the past research. At last, there is no public database for the new research direction. So, we build a new database, which consists of color image and depth map, to verify the usage of our method. The final experiments show that the depth information has a good effect and great significanceon on detecting the slianecy forground.