On the basis of the limitation of Farids universal blind steganalysis for color images, a new universal blind steganalysis scheme is proposed. In view of the luminance and the chrominance of the image, the class feature vector is extracted. A new approach is presented to extract the feature for the chrominance component that the image is directional processed using the vector directional filter, and then the vector direction correlation between the image and its directional filtered image is calculated. It reduces the number of the features enormously, and overcomes the fault of Farids scheme that neglects the inherent correlation between the RGB channels, which is propitious to simplify the design of classifier and to improve the performance of the classifier. Applying the scheme and Farids scheme to detecting the stego images produced by JSteg, F5, and EZStego respectively, the comparison of these simulation results shows that the scheme is quite more efficient for different size of messages embedded.