Pattern segmentation in textile images

This paper presents some results regarding the automatic segmentation of patterns in images of textile samples. The pattern segmentation is necessary for further extraction of a set of shape descriptors to be used in a CBIR database for textiles. Five different filters for removing the fabrics texture have been tested: mean, median, Gaussian, soft wavelet thresholding and a new approach based on wavelet transform and independent components analysis. For segmentation, due to the poor color content of the images, the binarization proved to be a satisfactory choice. Three algorithms for the threshold computation have been tested: Otsu, background symmetry and triangle.