A Novel Feature Fusion Technique in Saliency-Based Visual Attention

In this paper we proposed a novel feature fusion technique in Saliency-Based Visual Attention Model, presented in [Itti, 1998]. There are three conspicuity maps in Saliency-Based Visual Attention Model, which are linearly combined from 12 color maps, 6 intensity maps and 24 orientation maps (42 Feature maps overall) through an Across-scale combination and normalization. We utilized the genetic algorithm approach to combine all 42 Feature maps that are mentioned in this basic Saliency-Based Visual Attention Model. We proposed a "Weighted Feature Summation" to form a saliency map, with optimum weights which are determined by the genetic algorithm. The experimental results show the effectiveness of our proposed method to improve the detection speed of a favorite object in the scene.