Robust Saliency Map Comparison

ss_robust_metric MATLAB function performs comparison of saliency video sequence with ground-truth saliency video sequence using the method described in "Objective evaluation" section of

Y. Gitman, M. Erofeev, D. Vatolin, A. Bolshakov, A. Fedorov. "Semiautomatic Visual-Attention Modeling and Its Application to Video Compression". 2014 IEEE International Conference on Image Processing (ICIP). Paris, France, pp. 1105-1109.

The comparison carried out by the function is almost invariant to all common transformation, thus it allows to carry out fare comparisons of saliency models.

Features

• Metric is invariant to
  • Black and white level corrections
  • Gamma correction
  • Mixing method with Center Prior
• Finds the best center prior model for ground-truth sequence

Usage

function [ ss,per_frame_ss ] = ss_robust_metric(src_avi, gt_avi,sampling_density,optimization_iterations)

Input arguments:

• src_avi - path to AVI file with saliency sequence (uncompressed AVI recommended)
• gt_avi - path to AVI file with ground-truth saliency sequence (uncompressed AVI recommended)
• sampling_density - [DEFAULT VALUE = 400] sets frame sampling frequency for parameter optimization. Lower value leads to more precise results but increases computation time
• optimization_iterations - [DEFAULT VALUE = 100] set how many times method will try to find global extrema. Higer value leads to more precise results but increases computation time

Output arguments:

• ss - final metric value. Higher metric values stand for higher correlation with ground-truth
• per_frame_ss - vector of per-frame metric values

NOTE 1: First run of the the function with new ground-truth sequence takes longer than subsequent runs. On the first run method selects centre prior model for the ground-truth video. The selected ground-truth model will be cached for future used in *_cp.png file

NOTE 2: The method works really slow (it is the cost of its robustness), please, be patient

Usage example

File example.m contains usage example. It performs comparison of two saliency sequences and visualizes its results.

Misc

You can get ground-truth saliency sequencies from the project website. With any questions regarding this code usage you can contact Mikhail Erofeev [email protected]

If you use this code in your research please cite the following paper:

Y. Gitman, M. Erofeev, D. Vatolin, A. Bolshakov, A. Fedorov. "Semiautomatic Visual-Attention Modeling and Its Application to Video Compression". 2014 IEEE International Conference on Image Processing (ICIP). Paris, France, pp. 1105-1109.