A Saliency Detection Model Using Aggregation Degree of Color and Texture

Yan Wo,Xi Chen,Guoqiang Han

DOI: https://doi.org/10.1016/j.image.2014.10.004

IF: 3.453

2015-01-01

Signal Processing Image Communication

Abstract:Visual saliency detection aims to extract the region that attracts greater attention of human vision system than other parts of an image, which is widely exploited but remains a challenge in the computer vision field. Relative to the background, the salient regions are more aggregative in color or texture distribution. In this paper, we introduce the conception of aggregation degree, and propose a novel bottom-up saliency detection framework using the aggregation degree of color and texture. Firstly, the color saliency areas are extracted from each channel in the Lab and HSV color space, and the significance of each saliency area is evaluated by the aggregation degree and their divergence. Then the texture saliency map is extracted with the local phase of hypercomplex, which is constructed by the detail signal of stationary wavelet transform (SWT) and its Hilbert transform. Finally, we design an effective mechanism for credit feature maps selection depending on aggregation degree, and the final saliency map is reconstructed by fusing credit feature maps. Extensive experiments with ASD database and the performance comparison with seven state-of-the-art algorithms demonstrate that our algorithm achieves better performance.