An ELM based local topology preserving hashing
- PDF / 2,247,345 Bytes
- 18 Pages / 595.276 x 790.866 pts Page_size
- 107 Downloads / 163 Views
ORIGINAL ARTICLE
An ELM based local topology preserving hashing Yang Liu1,2 · Lin Feng1,2 · Shenglan Liu1,2 · Muxin Sun2,3 Received: 18 October 2017 / Accepted: 19 November 2018 © Springer-Verlag GmbH Germany, part of Springer Nature 2019
Abstract Hashing learning has become one of the most active research areas in computer vision and multimedia information retrieval with the explosively boosted data volume. Mainstream hashing methods adopt a two-stage hashing framework to realize hashing learning. That is, obtain low dimensional embedding and encode binary codes respectively. However, this kind of methods divides the dimensional reduction error and binary encoding loss apart, which is not beneficial to preserve the original data structure. Hence, we propose a local topology preserving hashing (LTPH) method to reduce the dimensional reduction error and binary encoding loss simultaneously. To clearly reveal the original data structure, Local Topology Preserving Embedding (LTPE) algorithm is proposed in this paper. LTPE utilizes the data similarity as well as the local geometry information to construct original data topology, which can effectively detect the original data structure. Nevertheless, LTPH is a transductive method, which is not suitable for large scale applications. Considering the outstanding global approximation ability and fast computation speed of Extreme Learning Machine (ELM), we propose an ELM based local topology preserving hashing (ELMLTPH) method to realize efficient hashing learning for large scale applications. With the facilitation of ELM, original data topology is effectively preserved to hamming space. Extensive image retrieval experiments are conducted on CIFAR, Caltech 101/256, Corel 10K and GIST-1M datasets, which demonstrate the superiority of ELMLTPH compared to several state-of-the-art hashing methods. Keywords Hashing learning · Extreme learning machine · Topology preserving · Large scale image retrieval
1 Introduction Large amount of data are accumulating with the rapid development of cloud computing, mobile internet and social internet [1–5]. The explosively boosted data arouses emerging * Lin Feng [email protected] Yang Liu [email protected] Shenglan Liu [email protected] Muxin Sun [email protected] 1
Faculty of Electronic Information and Electrical Engineering, Dalian University of Technology, Dalian 116024, Liaoning, China
2
School of Innovation and Entrepreneurship, Dalian University of Technology, Dalian 116024, Liaoning, China
3
State Key Laboratory of Software Architecture (Neusoft Corporation), Shenyang, China
need to data storage, indexing and analysis. Hashing learning methods compact original data by representing them with binary codes, which greatly save the storage cost and promote the effectiveness of indexing [6–9]. Since the learned binary representations can effectively and accurately facilitate large scale data analysis, hashing learning has attracted considerable attention both in industry and academia. Researches on has
Data Loading...
