Edge preserving image decomposition is proposed as a preprocessing technique to increase the accuracy in automatic radar target classification. The radar images are decomposed trough edge preserving image decomposition methods with parameters optimized for a better classification rate. By appropriate choice of the parameters, it is possible to keep the necessary information in the residual images while transferring the redundant information to the detail planes. Thus the use of residual image with the meaningful amount of data without redundancies increases the classification performance. The proposed preprocessing scheme is validated for an experimental dataset and compared with other image decomposition methods.