It has been shown that features can be selected adaptively for object tracking in changing environments [1]. We propose to use the variance of Mutual Information [2] for online fea...
A relevance filter is proposed which removes features based on the mutual information between class labels and features. It is proven that both feature independence and class condi...