In this paper a method which can improve the performance of face recognition system using color Gabor features is presented. First

quaternion is used to describe the color information

considering that Gabor filters have desirable characteristics of spatial locality and orientation selectivity

and they are extended to quaternion space. Then utilizing the convolution of the key points and the Gabor filters to extract features

by doing this the gray Gabor features are extended to the color ones. In the end

for the extracted features

we used PCA for dimension reduction and SVM for recognition. The experiment carried on Color FERET Database and the result utilizing ROC curve for cross-validation show that the use of color texture information can improve the efficiency of face recognition system markedly.