Color texture invariants for natural image recognition based on human visual system
J.F.C. Wanderley, Mark H. Fisher · 2002
The main contribution of this work is the proposal of a new set of color texture affine and illumination invariants based on the human visual system (HVS). A description regarding the importance of texture as a feature and a new color texture descriptor for natural image recognition, invariant to illumination and rotation, is introduced. The new feature vector extracted to represent the color texture images is generated by decomposing each of the three color channels of the image into several frequency and orientation bands using a bank of Gabor filters and then extracting several features from each color channel response independently and also from the combination of different color channel responses.