Clothing image retrieval based on a similarity evaluation method for Kansei retrieval system

Abstract

This paper proposes a similar image retrieval for the Kansei retrieval. The Kansei retrieval system is designed to search for user's clothing articles by considering their Kansei. The system adopts a Kansei retrieval agent. The agent identifies users's choice of clothing by imitating their Kansei. However, the agent's structure can become excessively complicated because considerable information is potentially available. Therefore, a Kansei retrieval system needs to utilizes similar image retrieval. Similar image retrieval searches for a clothing image on the basis of clothing images. Clothing design contains less information than clothing image. A design includes features intended to create a general impression of piece of clothing. In addition, the Kansei retrieval agent evaluates the design, and searches for the designs that a user likes on the basis of the evaluation. Therefore, the agent's structure does not become complicated. In this study, we created a similar image retrieval application and verified it effectiveness. The application searches for clothing images on the basis of the design that a user chooses. The results show that similar image retrieval can identify clothing images similar to the design. Thus, the effectiveness of the proposed similar image retrieval system was confirmed.