358 Feature Selection Combined Eye Tracking Data and Image Classification

Image classification often relies heavily on effective image descriptors. In this paper, a feature selection algorithm based on eye tracking data is proposed. This algorithm integrates the classification result based on support vector machines (SVMs) and mutual information difference (MID). In this method, regions of interest obtained based on the eye tracking data are used to represent the image. Then almost all low-level features collected are extracted for describing the above image regions. The SVMs classifier is used to perform a rough selection, while MID is used to obtain a smaller subset. Experimental results show significant improvement for feature selection by incorporating eye tracking data.