Visual clothing search by shape and style

Abstract

With a rapidly increasing use of the Internet, there is a large amount of digital image information, more and more picture information is applied to the commercial website. How to retrieve the digital image in database efficiently has been an significant issue. Besides described by words, pictures of clothes products are often attached. Traditionally, we tag each image with several keywords, so that users can search keywords to find products. However, the key words may contain no clue of what they are looking for. Clothes products often contain many characteristics which are hard to describe in regular keywords, such as texture, shape, or any other subtle details. To tackle this issue, we develop the image-based visual clothing search system, which enables users to retrieve clothing images. In this study, we provide the visual search with shape using the "Shape context" to be the shape descriptor so that one can search for similar shapes of clothes efficiently. We develop the system with two steps. First, appropriate feature description from clothes image is extracted. Second retrieval strategies of clothes image in database is built. We experiment on an image database with several kinds of clothes type and the experiment shows that our approach is more helpful for ranking clothes product by shape and style.