Abstract: A scheme for fabric defect detection based on Gabor filers is proposed in this paper. Based on the features of Gabor filers and textured materials with single color, a bank of ellipse-shaped Gabor filers with multi-scale and multi-orientation are designed. The fabric defect image is processed using this filter bank, and filtered sub-images are obtained which characterize the fabric defect in different orientations and scales in the frequency domain. These sub-images are fused in order to reconstruct the defect binary image that segments the defect from the texture background. Experiment results have confirmed the usefulness of this scheme for fabric defect detection.