Saliency detection based on spatio-frequency information
Shangwang Liu, Jianlan Hu, Yanmeng Cui · 2016
To increase accuracy of frequency-domain visual attention model, an improved Hyper complex Fourier Transform (HFT) algorithm is proposed. Firstly, the coefficients of the input hyper complex image were well designed. Real coefficient was adjusted intensity feature, and three imaginary coefficients were adjusted L, a, b color channels. Secondly, the 2-D signal was reconstructed by the original phase and filtered amplitude spectrum images, and a series of saliency maps were obtained. Finally, regarding the saliency maps as the density distribution function, the spatial contrast function was employed to generate the optimal saliency map. Experimental results show that the average AUC value of the proposed method is 0.8649, and F-measure achieves to 0.8274, which is better than related algorithms.