Salient object detection based on distribution-edge guidance and iterative Bayesian optimization
- PDF / 2,909,665 Bytes
- 14 Pages / 595.224 x 790.955 pts Page_size
- 40 Downloads / 200 Views
Salient object detection based on distribution-edge guidance and iterative Bayesian optimization Chenxing Xia1 · Xiuju Gao2 · Kuan-Ching Li3 · Qianjin Zhao1 · Shunxiang Zhang1
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Salient object detection has witnessed rapid progress, despite most existing methods still struggling in complex scenes, unfortunately. In this paper, we propose an efficient framework for salient object detection based on distribution-edge guidance and iterative Bayesian optimization. By considering color, spatial, and edge information, a discriminative metric is first constructed to measure the similarity between different regions. Next, boundary prior embedded with background scatter distribution is utilized to yield the boundary contrast map, and then a contour completeness map is derived through a wholly closed shape of the object. Finally, the above both maps are jointly integrated into an iterative Bayesian optimization framework to obtain the final saliency map. Results from an extensive number of experimentations demonstrate that the promising performance of the proposed algorithm against the state-of-the-art saliency detection methods in terms of different evaluation metrics on several benchmark datasets. Keywords Background scatter · Contour completeness · Iterative Bayesian · Salient object detection
1 Introduction Saliency detection is concentrated on detecting the most attractive objects in an image. Recently, this area has witnessed rapid progress. As a preprocessing procedure, automatic saliency detection has been widely used in a variety of computer vision tasks such as image segmentation [1], object recognition [2], compression [3], image retrieval [4], de-blurring [5], and others. Several saliency models have been proposed in the past years. Due to the lack of a uniform definition of salient objects, most salient object detection methods are based on effective assumptions. Contrast prior is one of the most popular principles adopted by various kinds of models from either a local or global view [12, 13]. Essentially,
Xiuju Gao
[email protected] 1
College of Computer Science and Engineering, Anhui University of Science and Technology, Huainan, China
2
College of Electrical and Information Engineering, Anhui University of Science and Technology, Huainan, China
3
Department of Computer Science and Information Engineering, Providence University, Taichung, Taiwan
local contrast-based methods [14] prefer to detect the highfrequency information such as edges, failing to pop out the salient holistic object, as shown in Fig. 1b. On the contrary, global contrast-based methods can locate the salient object while the performance of these methods is limited in such scenarios when the foreground regions are complex and with diverse appearance, as shown in Fig. 1c [13]. To address the limitation of the contrast cue, boundary prior is also applied to detect salient regions, where the image boundary areas are looked upon as background [15– 17]. For ex
Data Loading...
