A new rotation invariant feature extraction method for texture classification is proposed.2-D Fourier transform is applied on a texture image,a disk area within the central region of image is chosen,and frequency is sampled on the selected area with equal interval angles within the orientation［0°,180°］,so orientation decomposition is realized.A set of complex Morlet wavelet are applied on projection slice of each direction to decompose each projection into several frequency channels,the average and variance extracted are computed in each frequency channel,and then linear regression model is employed to computer realationship feature between frequency channels.1-D DFT is applied to features and the amplitudes of Fourier coefficient are selected as features,then the extracted features are rotation invariant.Experimental results show that features extracted have a good rotation invariant and better classification performance with some existing methods,and better classification results can also be achieved for non-rotation texture classification.