A new algorithm was proposed for extracting texture in complex multi-texture images. Firstly, polar coordinates method was used to analyze wedge and ring characters of texture frequency spectrum, and the period and directional parameters of texture distribution were computed. Secondly, based on these characters, ring and wedge Gauss pass-band filter banks were constructed to filter the frequency spectrum of multi-texture image. Thirdly, components of corresponding texture were acquired by transforming the filtered frequency spectrum image to time domain image. Lastly, through the rectification of texture area and corresponding post-processing, the true texture was extracted. The experimental results show that each texture is extracted truly and the elemental character is preserved exactly, and even in the area where distribution of texture is non-uniform, the skeleton of texture is still extracted truly.