A mass customization framework and reclassification method for lower garments in E-commerce
Ruibing Lin,Xiaoyu Lü,Pinghua Xu,Sumin Ge,Huazhou He
DOI: https://doi.org/10.1108/ijcst-04-2024-0096
2024-10-11
International Journal of Clothing Science and Technology
Abstract:Purpose To enhance the fit, comfort and overall satisfaction of lower body attire for online shoppers, this study introduces a reclassification method of the lower body profiles of young females in complex environments, which is used in the framework of remote clothing mass customization. Design/methodology/approach Frontal and lateral photographs were collected from 170 females prior, marked as size M. Employing a salient object detection algorithm suitable for complex backgrounds, precise segmentation of body profiles was achieved while refining the performance through transfer learning techniques. Subsequently, a skeletal detection algorithm was employed to delineate distinct human regions, from which 21 pivotal dimensional metrics were derived. These metrics underwent clustering procedures, thus establishing a systematic framework for categorizing the lower body shapes of young females. Building upon this foundation, a methodology for the body type combination across different body parts was proposed. This approach incorporated a frequency-based filtering mechanism to regulate the enumeration of body type combinations. The automated identification of body types was executed through a support vector machine (SVM) model, achieving an average accuracy exceeding 95% for each defined type. Findings Young females prior to being marked as the same lower garment size can be further subdivided based on their lower body types. Participants' torso types were classified into barrel-shaped, hip-convex and fat-accumulation types. Leg profile shapes were categorized into slender-elongated and short-stocky types. The frontal straightness of participants' legs was classified as X-shaped, I-shaped and O-shaped types, while the leg side straightness was categorized based on the knee hyperextended degree. The number of combinations can be controlled based on the frequency of occurrence of combinations of different body types. Originality/value This methodological advancement serves as a robust cornerstone for optimizing clothing sizing and enabling remote clothing mass customization in E-commerce, providing assistance for body type database and clothing size database management as well as strategies for establishing a comprehensive remote customization supply chain and on-demand production model.
materials science, textiles