Abstract:

Image saliency has been widely used in image segmentation, image retrieval, and image compression and so on. In order to solve the problems of the traditional algorithms, such as huge time consumption and sensitive to noise, we propose an improved multi-scale saliency detection based on hue, saturation, value (HSV) color space. The method chooses the hue, saturation and brightness of HSV color space as visual features. Firstly, we obtain the three-scale image sequences via the Gauss pyramid decomposition. And then, we extract each feature map from the three-scale sequence images through the improved SR algorithm. Finally, these feature maps are fused point to point by square operation and liner operation. Experiments show that, compared with the existing methods, the proposed method has better detecting effect and robustness, and it can quickly detect the saliency region of the image and highlight the entire salient object.

Key words: HSV color space, Gaussian multi-scale transform, spectral residual, salient map