The invention provides an image classification method; the method uses a depth decomposition scheme based on circular symmetry Gabor wavelet (CSGW), which is also called as DD-CSGW. For the CSGW has the rotation invariant property, the DD-CSGW is also the decomposition method of the rotation invariant. The depth decomposition refers to the iterated and layered decomposition through CSGW, and it means that a rough sub-belt can be continuously decomposed to be several more elaborated sub-belts through CSGW. For the DD-CSGW conversion domain is existed with strong dependence, we capture these dependences through a Copula model, namely, describe the decomposition sub-belt of every layer by Copula model. Meanwhile, the mean value and the variance of the Copula model parameter and the DD-CSGW sub-belt can be used for expressing the image, and the image classification is carried out by SVM. The depth decomposition method has good performance in image classification, and the classification stage speed is fast; besides, the method has the selection invariant property.