IEEE Trans. on Pattern Analysis and Machine Intelligence (PAMI) In press
If you have any question about this paper, please feel free to contact the first author (Jian Li). If you use this code, please cite our paper.
Some Experimental Results （see the complete results in the bottom)
The proposed model (HFT) is compared with SR,PQFT, Itti’s model, AIM, GBVS and Human labeled salient results.
Visual Saliency Based on Scale-Space Analysis in the Frequency Domain
Results of HFT
1: For each input image, HFT first compute 8 candidate saliency maps using Spectrum-Scale-Space in Hyper-complex Frequency Domain;
2: HFT choose one as the final saliency map output from these candidate saliency maps. See the choosing criterion in paper. As described in the paper, there are three versions of HFT: HFT*, HFT and HFT(e). The final saliency maps chosen by them are as follows:
Note that the number indicates that which one is chosen by the corresponding model from the candidate saliency maps.
Results of OTHER models