文章基本信息

标题：An Efficient Image Indexing Method Based on Spectral Hashing with Semantically Consistent Sparcified Graph Laplacian Approach
本地全文：下载
作者：M.Premesudha ; P.Kavitha
期刊名称：International Journal of Computer Trends and Technology
电子版ISSN：2231-2803
出版年度：2014
卷号：15
期号：1
页码：27-30
DOI：10.14445/22312803/IJCTT-V15P106
出版社：Seventh Sense Research Group
摘要：Similarity search is one of the most primary problems in information retrieval, and machine learning research groups. The ability of rapid similarity search in a largescale dataset is of huge significance to several multimedia applications. Semantic hashing (SH) is a challenging way to accelerate similarity search, which plans flexible binary codes for a huge number of images with the intention that semantically similar images are planned to close codes. Earlier work used Spectral hashing with pair wise similarity in the hash function learning process. This can be optimized by graph Laplacian. For large number of image datasets, the method will not broaden its performance as like the performance shown in small image dataset. In order to overcome from this problem, optimized informative graph has been learned and constructed. In this paper Graph Laplacian with Sparcification has been proposed for obtaining optimized informative graph. This method directly optimizes Laplacian graph. The learned graph with sparcification represents improved similarity between samples, is then applied to Spectral Hashing for generation of efficient binary codes. Unlike metric learning, sparcified Laplacian graph automatically determines the scale factor during the optimization. Experimental result of proposed method on publicly available datasets achieves better result when compared with earlier works.
关键词：Sparcification; Graph Laplacian; similarity search; spectral hashing