CAJ | 학술논문

基于哈希编码的算法，由于其高效性，已经成为海量数据高维特征最近邻搜索的研究热点。目前存在的普遍问题是，当哈希编码长度较低时，原始特征信息保留不是很充分，从而导致检索结果不理想。为了解决这一问题，提出了一种基于Markov网络的有效哈希编码算法。该算法首先根据稀疏编码策略进行特征重构，通过Markov随机游走的方式构建特征之间的语义网络关系图，然后根据Laplacian特征映射求出投影函数，最后进行快速的线性投影二值化编码。在公开数据集上与主流算法进行了性能比较，实验结果表明该算法具备良好的检索性能。
기우합희편마적산법，유우기고효성，이경성위해량수거고유특정최근린수색적연구열점。목전존재적보편문제시，당합희편마장도교저시，원시특정신식보류불시흔충분，종이도치검색결과불이상。위료해결저일문제，제출료일충기우Markov망락적유효합희편마산법。해산법수선근거희소편마책략진행특정중구，통과Markov수궤유주적방식구건특정지간적어의망락관계도，연후근거Laplacian특정영사구출투영함수，최후진행쾌속적선성투영이치화편마。재공개수거집상여주류산법진행료성능비교，실험결과표명해산법구비량호적검색성능。
Hashing coding is becoming increasingly popular for efficient nearest neighbor search in massive database, due to its high efficiency. However, the exiting hashing methods cannot achieve a satisfied performance with low hash bits because of less original information. To address this problem, this paper proposes a novel method called hashing with Markov random walk graph. Firstly, the method generates sparse representation for high dimensional vectors, then builds the semantic network between the sparse coding by Markov random walk. After that the projec-tion function can be found by Laplacian eigenmap method and the binary hashing coding can be generated by linear projection quickly. Experimental comparisons on two large datasets demonstrate that the proposed method outper-forms the state-of-the-art methods.