A New 2D Optimal Digital Filter（ODF）which is used to extract texture features of the image is introduced in this paper. The design of the filter is based on the finite prolate spheroidal sequences （FPSS） and the discrete fourier transform （DFT）.The frequency response of the filter is optimal in terms of energy loss. A distribution of the filters with different center frequency and band width over the frequency plane is used to extract texture features of the image then proceed to classify and to segment of them by using a nonparametric classifer. The practical segmented examples of an artificial and a natural texture image are shown in the paper.