Multiscale Discriminant Saliency for Visual Attention

TL;DR

This paper introduces a multiscale discriminant saliency method using wavelet features and Hidden Markov Trees to improve visual attention modeling, validated through quantitative and qualitative evaluations against existing methods.

Contribution

It presents a novel multiscale saliency detection approach combining wavelet-based features and probabilistic modeling for enhanced visual attention prediction.

Findings

MDIS outperforms AIM in quantitative metrics

The method effectively captures multiscale saliency features

Simulation results verify the validity of the approach

Abstract

The bottom-up saliency, an early stage of humans' visual attention, can be considered as a binary classification problem between center and surround classes. Discriminant power of features for the classification is measured as mutual information between features and two classes distribution. The estimated discrepancy of two feature classes very much depends on considered scale levels; then, multi-scale structure and discriminant power are integrated by employing discrete wavelet features and Hidden markov tree (HMT). With wavelet coefficients and Hidden Markov Tree parameters, quad-tree like label structures are constructed and utilized in maximum a posterior probability (MAP) of hidden class variables at corresponding dyadic sub-squares. Then, saliency value for each dyadic square at each scale level is computed with discriminant power principle and the MAP. Finally, across multiple scales is integrated the final saliency map by an information maximization rule. Both standard quantitative tools such as NSS, LCC, AUC and qualitative assessments are used for evaluating the proposed multiscale discriminant saliency method (MDIS) against the well-know information-based saliency method AIM on its Bruce Database wity eye-tracking data. Simulation results are presented and analyzed to verify the validity of MDIS as well as point out its disadvantages for further research direction.