一种改进的频域视觉模型.docVIP

Saliency detection based on frequency domain analysis and spatial information Shangwang Liu1, 2, Jianlan Hu1, 2 1. Henan Normal University, College of Computer and Information Engineering, No. 46, East Jianshe Road, Xinxiang 453007, Henan Province, China 2. Engineering Lab of Intelligence Business Internet of Things, Henan Province, China Abstract- Saliency detection acts as a vital role for a variety of computer vision applications, such as image segmentation, target recognition, image recognition and so on. Based on the proven biological visual attention mechanism, the paper researches on simulating and simplifying its computability theory and proposes a novel saliency detection model with frequency domain analysis and spatial information. Firstly, based on the quaternion representation for image, the real and imaginary coefficients of the quaternion has been chosen, the coefficient of real part is the linear fusion of direction and intensity, while three imaginary part coefficients are H, S, V three color components respectively in improved HSV color space. Secondly, after analyzing the spectrum scale space in frequency domain, a series of salient maps at different scales have been obtained. Finally, in order to avoid highlighting the edge of salient object in the frequency domain processing and the redundant background, the spatial standard deviation has been used to select the optimal saliency map. In the biggest database with salient objects labeled precisely, the qualitative and quantitative comparison experiments of 4 kinds of classic methods have been made, the averages of AUC of the proposed method is 0.8454, and F-values gets 0.8300, and the results show that our method clearly outperforms alternate, state-of-the-art algorithms. Index Term- frequency domain visual attention model; spatial information; quaternion; standard deviation; saliency map 0 INTRODUCTIONS In recent years, with the popularity of electronic products and the rapid developmen