Abstract: Focusing on objective assessment of fabric pilling grading, a method was proposed for detecting pilling object using scale-space extrema. The pilling object was modeled as an anisotropic Gaussian kernel. Based on scale-space theory and derivation of isotropic Gaussian matched filter, an operator as polynomial combinations of Gaussian derivatives was used for automatic scale selection, which providing a close approximation to Gaussian matched filter. By scale-space extrema of the normalized operator filtering , the pilling object was located and its size was measured. The anisotropic Gaussian model parameters were estimated from local structure tensor matrix, and depending on them, the pilling object was finally segmented and recognized. The experiments demonstrated that the presented method of pilling object segmentation and recognition has good results.

Key words: fabric, pilling, scale-space theory, object detection