Salient Region Detection by Region Color Contrast and Connectivity Prior
The visual salient regions detection is one of the fundamental problems in computer vision, so saliency estimation has become a valuable tool in image processing. In this paper, we propose a novel method to realize the calculation of saliency, using color
- PDF / 641,082 Bytes
- 10 Pages / 439.37 x 666.14 pts Page_size
- 44 Downloads / 227 Views
College of Information Science and Engineering, Yanshan University, Qinhuangdao 066004, Hebei, China [email protected] 2 Hebei Province Key Laboratory of Computer Virtual Technology and System Integration, Qinhuangdao 066004, China
Abstract. The visual salient regions detection is one of the fundamental problems in computer vision, so saliency estimation has become a valuable tool in image processing. In this paper, we propose a novel method to realize the calculation of saliency, using color contrast and connectivity prior (called CCP for short). There are three cues integrated to obtain high-quality map, including contrast, spatial distribution and high-level prior. We evaluate our approach on three standard benchmark datasets with other state-of-the-art approaches, the results show that the proposed method has the higher precision and recall, the final maps are more closed to the ground truth. Keywords: Salient region detection · Contrast · Spatial distribution · Connectivity prior
1
Introduction
People can easily focus attention on the important parts or the salient regions in a scene. Salient object detection has been studied in physiology, neural systems, psychology and computer vision for a long time. It is motivated by the importance of saliency detection in applications such as image segmentation [1,2], object recognition [3], image retrieval [4,5], adaptive compression of images [6] and so on. Nowadays, existing visual attention approaches mainly include two kinds, namely, fast, bottom-up, data driven saliency geodesic; and slower, top-down, task driven saliency geodesic. The former is popular, and some models are very successful in salient region detection [7-15], which almost focus on the contrast of low level image features. While the contrast information often helps produce good significant results, it always generates high values for the area which is not salient, especially for the regions with low contrast from the surrounding or connected heavily with the object. This research is partly supported by the National Science Foundation, China (no. 61379065). Hebei Province Science and Technology Support Program, China (no. 13211801D). The Doctoral Foundation of Yanshan University (no. B540). © Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part II, CCIS 547, pp. 21–30, 2015. DOI: 10.1007/978-3-662-48570-5_3
22
M.-H. Chen et al.
Inspired by the above diiscussion, in this paper, we propose a new algorithm off the contrast, i.e., region color contrast. Our method is based on the assumption that the salient region is not only assigned high contrast values, but also located near the image center, besides, warm ccolors are more pronounced. The contrast may work w well for low-level saliency calculation, but they are neitther abundant nor high-accuratee if used alone. So several approaches [16-26] exploited some high-level prior know wledge to help get more sufficient values. [19-21] made use of the boundary prior (or caalled connectivity prior). These articles simply thought th
Data Loading...
