Quantifying the contribution of feature maps for goal-directed visual attention

Assessing and selecting relevant visual cues is crucial for rapid saliency estimation and visual search. Here, we derive a new optimal feature modulation strategy to maximize the relative salience of the target, in which the top-down weight on a feature map depends on its stimulation intensity ratio (SIR) between the target and the distractors. The stimulation intensity is determined by two factors, i.e., cumulative summation of salience and the mean activity coefficient. Furthermore, we present a pruning strategy (i.e., extracting a small subset of features to compute the corresponding feature maps whose weights are higher than a given threshold prior to the feature combination) to reduce the computational cost of the search process. Testing on natural scenes shows that our optimal feature gain setting strategy together with the pruning technique increase the search speed and accuracy.