Scale-space saliency detection in combined color space

Interesting objects can appear more salient either in the RGB color space or in the Lab color space depending on the input images case by case. With this observation, saliency detection in a combined color space is proposed and investigated in this paper. In particular, the adaptively-centered Hypercomplex Fourier transform (A-HFT) algorithm in the RGB color space proposed in a recent work is performed for saliency detection in the combined color space. First, the input image is transformed into the frequency domain by the Hypercomplex Fourier transform in terms of its RGB and Lab representations, respectively. Next, a set of low-pass Gaussian kernels are employed to perform a scale-space analysis, resulting in a series of RGB and Lab saliency maps. Finally, the optimal saliency maps in the two color spaces are determined by minimum entropy, and the better one (which has lower entropy) is taken as the outcome of the combined color space. The efficiency of the new algorithm is evaluated by extensive experimental results.