Visual Attention Model Based on Multi-Scale Local Contrast of Low-Level Features

Salient regions detection is becoming more and more important due to its useful application in image representation and understanding. The accurate detection of salient regions can reduce the complexity and improve the efficiency of image processing. In this paper, a visual attention model based on multi-scale local contrast of low level features is proposed. In the proposed model, a multi-scale transform is used to obtain the original image at different scales, and the local contrast features of intensity, texture and color are calculated at each scale. Then these contrast features are interpolated iteratively to form three feature maps corresponding to intensity, texture and color respectively. Finally, the feature maps are integrated to obtain the final salient regions. In the experiment, a proven eye tracking system is used and verifies the salient region detected by the proposed model consistent with human vision. Furthermore, comparing with another two existing models, the proposed model also shows better performance.