A novel feature weighting method for FIT Based saliency area detection

In this paper, a novel method is proposed to weight the contribution of different feature maps for Feature Integration Theory (FIT) based saliency area detection in static images, aiming to imitate the bottom-up attention mechanism of Human Visual System. Firstly, the same topographic representation feature maps are generated through extracting color, orientation features of an original image. Then the corresponding weight is computed for each feature map based on its shape parameter of Generalized Gaussian Distribution (GGD) and variance of position distribution so as to generate the saliency map through combination of feature maps with their computed weights. Through comparison with experimental eye tracking data, the results on natural images demonstrate that our proposed method is more efficient than other existing methods on jointly consideration of predicting accuracy and time consuming. Moreover, this method is also fairly universal and can be applied to the other existing FIT based saliency computational models.