Abstract:

As the effect of the variety of fabric background textures and the feature of pilling defects, conventional algorithm of image processing is hard to satisfy the automatic detection of pilling defects and the objective evaluating demands. A new way of pilling objective evaluation based on the wavelet-domain of Difference of Gaussian  filter was proposed. First of all, pilling defect image was decomposed into multiple layers by wavelet multi-decomposition to separation periodic background texture and pilling information. Then, the appropriate wavelet decomposition sub-images were chosen to carry on difference of Gaussian filter for eliminating the noise and the background information of slow variation such as uneven illumination, and pilling information was improved significantly; and on this basis, a segmentation threshold was defined to segment these sub-images according to the characteristics of pilling, and the features of pilling form binary image was extracted. Finally, BP artificial neutral network wasused to objective evaluation of pilling grade. The test results show that this method can make an objective evaluation for pilling level effectively, and has strong interference resistance.

Key words: pilling, defect, image, wavelet domain, difference of Gaussian filter, threshold segmentation, objective evaluation