Human vision is gradually substituted by computer vision in the field of products surface detection. This is the inevitable trend of industrial automation control. However, surface texture is always a difficult point for defects detection. This paper starts from the angle of image segmentation, treating defects as objects and textures as backgrounds, to extract the surface defects. Based on the non-parametric statistical activie contour model, a distribution prior model is proposed. Our model utilizes the distribution of textures as the prior distribution of the background, which makes it easier to distinguish the defects and the textures. Experiment results show that the proposed method adapts to defects detection in different texture background, and exactly extracts the location of defects.