Objective automatic assessment of pilling in fabrics by image analysis

Abstract

A fully automatic method for pilling evaluation in fabrics, pre- viously developed from the analysis of Zweigle standard images, is com- pleted for its application to the assessment of real samples. Several aspects are dealt with. As regards the acquisition of input images, a set of primary images sequentially captured under near-tangential illumina- tion from different sides is proposed to obtain complete information of the pill area along with a good pill-to-background contrast. The significant information for the set is concentrated in a single image using the Karhunen-Loeve transform in a later stage of the algorithm. After seg- mentation, pills appear segmented from the background, and binary noise is reduced by applying a threshold of minimum pill area. This threshold is scale independent when it is expressed in terms of the fab- ric's thread crossings. From the final segmented binary image, two pill features are measured: the total area of pilling, from which the degree of pilling is assigned, and the pill density. The complete method is applied to 32 real samples, which constitute a wide variety of cases. Three labo- ratories with their respective panels of experts for visual rating are in- volved in the assessment of the samples. The results obtained by the image-analysis method and by the visual raters are compared. Finally, the robustness of the method against slight changes in some parameters of the algorithm is evaluated and compared with the degree of agree- ment of experts. © 2000 Society of Photo-Optical Instrumentation Engineers. (S0091-3286(00)01806-7)