In this paper we propose three types of features which can be used to realize efficient universal steganalysis on color images. After measuring the deviation of DCT coefficient distribution

the correlation of smoothness in spatial domain

and the correlation between different components

a 10 dimentional feature vector is produced for each color image. Those features are sensitive to steganographic process and the feature vector

having a low dimension

is easy to calculate. Then

a SVM classifier is built to distinguish the stego images from the original images. Experiments show that the universal steganalytic approach can effectively detect the presence of secret message hidden by Jsteg

F5 or MB steganography

and can identify the steganographic techniques used.