Digital 3D system for classifying fabric pilling based on improved active contours and neural network

In this paper, we proposed an objective 3D evaluation system of pilling. It can be applied to fabrics with intricate patterns and textures. The major features of the pilling evaluation system include a 3D fabric surface reconstruction based on a multi-view stereo vision algorithm, an improved active contours model for pill segmentation, and objective pilling grade calculation utilizing a neural network. To improve the accuracy of pilling grade estimation, the obtained maxima algorithm is used to locate the pilling, which is then used as the seed point for curve evolution until the curve converges to the pilling's edge. In addition, the binarization and adaptive threshold methods can also be utilized to obtain the pilling binary graph. Three feature parameters, such as the pill number, area, and coverage are extracted from the segmented binary graph, which is used to objectively evaluate the pilling grade of fabric. The experimental results show that the objective evaluation system can accurately evaluate the fabric pilling grade, which is highly consistent with the subjective evaluation results.