Robust feature extraction for material image retrieval in fashion accessory management

Abstract

Fashion accessory plays an important role in costume designing. A well-designed accessory consisting of different types of materials help enhance the aesthetic of the dresses. A key problem of accessory design is to find the replaceable material with appropriate aesthetic and cheaper price. However, such a process is performed manually in accessory factory, in which the work efficiency is very low. Therefore, material image retrieval is an important technique to automatic and facilitates the process of accessory design and management. In this paper, a voting-based preprocessing method is proposed to locate the material in the image. And thus a regression model is built to make use of the neighboring edge directions to optimize the robust edge direction of a point. Finally, both color and edge features will be coded as histogram-based features for representing the materials for image retrieval. Experiments have been conducted on real captured material image to validate the effectiveness of the proposed locating and searching technique.