Visual saliency detection by integrating spatial position prior of object with background cues. (15th April 2021)

• Record Type: Journal Article
• Authors: Jian, Muwei; Wang, Jing; Yu, Hui; Wang, Guodong; Meng, Xianjing; Yang, Lu; Dong, Junyu; Yin, Yilong
• Is Part Of: Expert systems with applications. Volume 168(2021)
• Publication Date: 2021-04-15
• Subjects: Discrete wavelet transform -- Saliency detection -- Background features -- Position prior; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2020.114219
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: In this paper, we propose an effective visual saliency-detection model based on spatial position prior of attractive objects and sparse background features. Firstly, since multi-orientation features are among the key visual stimuli in the human visual system (HVS) to perceive object spatial information, discrete wavelet frame transform (DWDT) is applied to extract directionality characteristics for calculating the centoid of remarkable objects in the original image. Then, the color contrast feature is used to represent the physical characteristics of salient objects. Thirdly, in order to explore and utilize the background features of an input image, sparse dictionary learning is performed to statistically analyze and estimate the background feature map. Finally, three distinctive cues of the directional feature including the color contrast feature and the background feature are combined to generate a final robust saliency map. Experimental results on three widely used image datasets show that our proposed method is effective and efficient, and is superior to other state-of-the-art saliency-detection models. Highlights: We design a weighted centroid calculation algorithm for center prior generation. We explore the 'forward' and 'backward' visual cues for saliency detection. A backward- and two forward- visual cues are integrated to generate saliency maps.