Tensor Based Feature Detection for Color Images

Extending differential-based operations to color images is hindered by the multi-channel nature of color images. The derivatives in different channels can point in opposite directions, hence cancellation might occur by simple addition. The solution to this problem is given by the structure tensor for which opposing vectors reinforce each other. We review the set of existing tensor based features which are applied on luminance images and show how to expand them to the color domain. We combine feature detectors with photometric invariance theory to construct invariant features. Experiments show that color features perform better than luminance based features and that the additional photometric information is useful to discriminate between different physical causes of features.