Fabric image segmentation is very important for fabric designing, fabric manufacturing and textile printing and dyeing. The results of fabric image segmentation are inevitably influenced by some noisy factors, such as fibers, dust, stains, holes, neps, and other fabric defects. In this paper, a novel fabric image segmentation method based on rough possibilistic clustering is presented. Specifically, all pixels are partitioned into three approximation regions with respect to a fixed cluster, i.e., the core, boundary, and exclusive regions of this cluster. The corresponding prototype calculation is only related to the core and boundary regions, rather than all the image pixels. Thus, the obtained prototypes cannot be distorted by the pixels belonging to other colors or regions. In addition, since the typicality values are involved, the proposed method is robust for dealing with noisy environments. The improved performance of the proposed method is illustrated by some real fabric images.
Rough possibilistic clustering Approximation region Uncertainty Fabric image segmentation
This is a preview of subscription content, log in to check access.
The work is supported by Postdoctoral Science Foundation of China (No. 2017M612736, 2017T100645), Guangdong Natural Science Foundation (No. 2018A030310450, 2018A030310451).