Indexing and Retrieval of Textile-Curtain Database Using Wavelet Domain Features

概要

論文の詳細を見る
Handling a large amount of images in textile and clothing industry is an important task, which calls for the development of an effective retrieval system. However, a limited number of works have been observed in the literature. Most of which emphasizes on color disregarding typical basic patterns which are perceptually more meaningful. Some typical patterns are horizontal, vertical, cross stripes, leaves and flowers. In this study, we propose an indexing and retrieval system based on wavelet domain features namely Directionality, Regularity and Symmetry, which uses correlation and edge information of the wavelet subbands in the two major axes (horizontal and vertical). In order to reduce searching time, we first categorized various patterns using supervised LVQ neural network. Then group representative prototypes are formed averaging all feature vectors in a particular group. For a typical query, the query key is first compared with a few group prototypes to determine the expected category. Then it undergoes comparison within that particular group and retrieves relevant images. User can also interact to obtain correct category. An experiment with a set of curtain images shows the potentiality of the proposed approach compared to conventional wavelet energy feature.
社団法人電子情報通信学会の論文
2004-01-15