Research On Topology Preserving Hashing Method

With the development of the Internet and electronic devices, data, e.g., texts, images and videos are growing rapidly. For many scenarios, people need to retrieve relevant content from such large scale data. However, for large scale data, it is computationally infeasible to find the closest points of a query in a given database. To tackle this problem, recently, a lot of efforts have been devoted to approximate nearest neighbor (ANN) search, which returns approximate nearest neighbors instead of exact nearest neighbor, but is fast.Hashing based ANN search is a classical method for approximate nearest neighbor search. Generally, such methods first map data points into hamming space; then, in hamming space, they make comparison. They have efficient constant query time and can also reduce storage by storing compact codes of data points. Thus, in these years, hashing based methods have attracted more attention from both academia and industry areas.In this thesis, we propose a new hashing method-Self-organizing map hashing(SOMH). In SOMH, an self-organizing map is used to map original data into hamming space. Self-organizing map can keep the topology structure while doing the mapping. It is reasonable to make use of it in hashing methods. In addition, for long and short hashing codes, we propose different schemes to adjust the proposed method to obtain good performance.To verify the performance of the proposed method, we test it on different datasets. From the experimental results, we find that SOMH outperforms or is comparable to some state-of-the-arts.