Based on wavelet coefficient dependency,a novel general image steganalysis technique for spatial domain steganography is proposed.First,the mutual information is exploited to analyze the change on the scale and orientation dependency between wavelet coefficients,which is caused by the embedding of a message,and the Markov model is applied to model the dependency between wavelet cofficients so as to extract intrascale and interscale transition probability matrices which are used as feature vectors.Then,a weighted feature fusion method is used to fuse these feature vectors and the fisher linear discriminant(FLD) is designed to classify them.The experiments on least significant bit(LSB),LSB matching and stochastic modulation(SM) steganography show that the method can detect stego images reliably and the detection accuracy of this method exceeds that of its closest competitors obviously under the same computer complexity.