文献类型：期刊文献

英文题名：Reversed Sketch: A scalable and comparable shape representation

作者：Huang, Ming[1];Lin, JiaJun[1];Chen, Ning[2];An, Wei[3];Zhu, WeiJian[4]

机构：[1]East China Univ Sci & Technol, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China;[3]Chinese Acad Sci, Inst Informat Engn, State Key Lab Informat Secur, Beijing 100093, Peoples R China;[4]Chinese Acad Sci, State Key Lab Network Data Sci & Technol, Beijing 100093, Peoples R China

年份：2018

卷号：80

起止页码：168

外文期刊名：PATTERN RECOGNITION

收录：;EI(收录号：20181304962983);WOS:【SCI-EXPANDED(收录号:WOS:000432511200014)】；

基金：This work was supported in part by Shanghai Public Security Bureau and by Shanghai Municipal People's Government under grant No. 31010600000000020141A2101001 and by National Natural Science Foundation of China under grant nos. 61771196 and 61402436.

语种：英文

外文关键词：Shape representation; Image matching; Contour detection; Polygon evolution; Content-based image retrieval

摘要：The shape features of images are essential to image recognition, comparison and retrieval since most users are more interested in recognizing or comparing images by shape than by color and texture [1]. Comparing or retrieving images by shape is still envisioned as one of the most challenging works in image comparison and retrieval because of the lack of effective and efficient representations of shape features in image comparison and retrieval. In this paper, we propose a scalable and comparable shape representation, namely "Reversed Sketch", is proposed, on which a shape feature extraction and utilization framework is built. With this representation, we represent an image object using a polygon extracted from the contour of the image object by a force-driving sliding box algorithm. A polygon evolution algorithm is then proposed for transforming the first wiggly polygon into a more sketchy form for efficiently processing, which makes our shape representation more scalable. Also, we present a comparable metric drawn from this representation combined with the comparing algorithm, which is invariant to scaling, rotation and translation and thereby is suitable for image recognition, registration and comparison. The proposed shape representation is especially suitable for image retrieval because with it a hierarchical index which is very useful for image retrieval can be built on the image dataset. Extensive experiments are carried out and the experiment results show that with our shape representation the shape features can be quickly extracted from an image, simplified as needed, and used to efficiently comparing shapes in accord with people's perception. Experiment results derived with practical datasets indicate that our framework can achieve better comparing precision and efficiency, compared with some other methods. (C) 2018 Elsevier Ltd. All rights reserved.