A New Approach Based on Ensemble Clustering for the Fabric Color Batching Problem

The fashion industry is one of the industries most influenced by aesthetics and quality. This necessitates that products manufactured for this industry possess high quality and aesthetic appeal. Denim products are among the most frequently used in this industry for various purposes. This study proposes an ensemble clustering approach for visually sorting batches to reliably classify color consistency in denim fabrics. First, separate batches were obtained using three common methods (DBSCAN, hierarchical clustering, and K-Means) with 800×800 pixel RGB images of fabric samples for each order. Then, an ensemble rule based on the majority principle was designed to reduce inconsistencies between methods and balance random initialization and parameter sensitivity. Each sample was assigned to the final batch according to the majority preference among the batches given by the three algorithms. It is evaluated that the proposed approach by comparing it with reference batch assignments predefined by experts. The outputs of the individual algorithms and the ensemble results are compared each other. The findings show that the ensemble rule produces more stable batches that are closer to expert decisions. While preserving the strengths of the individual methods, the ensemble rule reduces the impact of their weaknesses.