Color textural features under varying illumination

In this paper, we present a new feature extraction methodology for color texture recognition. It is based on the covariance of 2^nd-order statistical features in the wavelet domain of the color channels of the images and it is named as color wavelet covariance (CWC). The experimentation showed that the CWC features could be used effectively for texture representation even when illumination varies. The use of the linear K-L (Karhunen-Loeve) transformation of the RGB color space for the extraction of the CWC features resulted in a performance that was comparable to the one achieved with more complex non-linear color transformations. The recognition accuracy tested with texture mosaics reached an average of 86%. Using images acquired under varying illumination the performance of the CWC features on the K-L space reached an average of 88%.