A form attribute classification system for fashion items

Abstract

ABSTRACTMachine learning generates rules by learning data; therefore, the reliability of the training data is one of the important factors of learning accuracy. In the field of fashion, it is important to systematically label fashion items using attributes in order to achieve an image retrieval service using machine learning. Thus, the aim of this study is to construct a fashion attribute hierarchical classification system. To do this, the meta-data of fashion item image collected from the consumer was analysed and professional fashion literature was referenced. This study proposes a hierarchical classification system that classifies dimensions and attribute-values that satisfy consistency, exclusiveness, inclusiveness, and flexibility by combining meta-data and professional fashion literature. This will improve the reliability of the training data, which is an essential element in machine learning, by providing standardised criteria for tasks such as tagging and labelling of fashion items.