Reduced quaternion matrix for color texture classification

As a newly developed singular value decomposition of the reduced quaternion matrix (SVDRQ), the two reduced quaternion unitary matrices can effectively capture the intrinsic geometric structures and smooth contours of color texture image. The projection vector by the two unitary matrices is used as dominant features for color texture classification. In this paper, we proposed new algorithm to implement the computation of the SVDRQ, and then proposed new color texture classification scheme based on SVDRQ, the Euclidean distance is applied as classifier in the proposed scheme. It is demonstrated by the experiments that our proposed scheme significantly improves the color texture classification accuracy in comparison with several conventional texture classification approaches.