In view of the problems in the textile process, a novel fabric defect segmentation method-quartering method and a fabric defect feature extraction method-Radon Wavelet Low Resolution Characteristic (RWLRC) was presented, which were respectively used for fabric defect detection and classification. According to this method, the fabric image was preprocessed by using Gabor filter, and then the fabric image was divided into four parts, the threshold for segmenting the fabric defect was determined by four parts' maximum value and minimum value. After that the Radon transform was used to binary image and characteristic curve was got. Meanwhile Mallat pyramidal decomposition algorithm was used for feature dimension reduction. Finally, the neural network was used to the state recognition and characteristic classification. The experimental results show that quartering method does not need to contrast with the other normal fabric images and has good adaptability. RWLRC only has three eigenvalues and has the characteristics of low dimension and accurate description of defect shape, the proposed method can efficiently inspect and recognize four common fabric defects:weft-lacking, warp-lacking, oil stains and holes.