The detection of embedded information for spatial type of information hiding methods is easy, if the embedding algorithm is known. Also detection shame can be efficient if it is specific for embedding methods. However, in this case always the detector is step behind the embedding technique. With this idea in mind we develop two approaches that are universal blind detector, and RS detector. The universal blind detector is based on a collocating image feature from clear image, and stego image (image embedded information). These features based on two multi-scale image decompositions, the quadrature mirror filter (QMF) pyramid decomposition and the local angular harmonic decomposition. The classifier is used to classifying the clear from stego image. We propose three types of classifiers that are linear discriminant analysis, non-linear support vector machines, and one class support vector machines. The experiments confirm that the proposed image statistics and nonlinear classification are effective in generic image steganalysis. Also RS steg analysis method is developed for general LSB embedded methods. In which an accurately estimation of the number of changes to the cover image imposed during embedding for random straddling for palette images, and the classical LSB embedding. The experiments confirm that; the developed RS method is around 10 times better than the old RS steganalysis methods. With that increment in accuracy the execution is also reduced.