Saliency detection by combining spatial and spectral information.

In this Letter, a new algorithm is proposed to detect salient regions by combining spatial and spectral information. First, the input image is considered in both RGB color space and Lab color space. Second, the biggest symmetric surround model and spectral residual are calculated in each channel simultaneously. Third, the feature maps in some color channels outperform the feature maps in the other channels. Entropy is defined to evaluate the performance of the feature maps, which can be used to choose the proper channels and combine different feature maps. Finally, a Gaussian low-pass filter is applied to improve the performance by accounting for the center bias. Compared with previous methods, our saliency detection is more effective and robust as demonstrated by the experiments.