1.确定检索范围
我需要找一些有关Hash在图像检索方面的应用相关的paper,即怎么在编码时保证图像在原始特征空间的相似性。
2.文献搜索
使用Google学术进行搜索。搜索的关键词为“Graph”,“Hashing”,时间选择为2015年以来,搜索结果如下。
3.浏览标题和摘要
可以看到,“Scalablegraph hashing with feature transformation”的引用次数较多,所以我选择先看这一篇的摘要。摘要的大致内容是他们提出了Scalable Graph Hashing(SGH)的方法,通过特征转换对原图像进行哈希编码,哈希编码要保留原图像间的相似性,他们的方法在两个大数据集上取得了很好的效果。这与我想找的paper内容相一致。
通过此方法找到了其他几篇paper分别是:
[1]Guosheng Lin,Chunhua Shen, Qinfeng Shi, Anton van den Hengel, and David Suter. Fastsupervised hashing with decision trees for high-dimensional data. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition,pages 1971–1978, 2014.
[2]Dongqing Zhangand Wu-Jun Li. Large-scale supervised multimodal hashing with semanticcorrelation maximization. In Proceedings of the AAAI Conference on ArtificialIntelligence, pages 2177–2183,2014.
[3]Alexandr Andoniand Piotr Indyk. Near-optimal hashing algorithms for approximate nearest neighborin high dimensions. Communications on ACM,51(1):117–122, 2008.
[4]Yunchao Gong andSvetlana Lazebnik. Iterative quantization: A procrustean approach to learningbinary codes. In Proceedings of the IEEE Conference on Computer Vision andPattern Recognition,pages 817–824, 2011.
[5]Mark J. Huiskes,B. Thomee, and Michael S. Lew. New trends and ideas in visual conceptdetection: The mir flickr retrieval evaluation initiative.In Proceedings of theACM International Conference on Multimedia Information Retrieval, pages527–536, 2010.
[6]Wei Liu, JunWang, Sanjiv Kumar, and Shih-Fu Chang. Hashing with graphs. In Proceedings of theInternational Conference on Machine Learning, 2011.
[7]Yair Weiss,Antonio Torralba, and Robert Fergus. Spectral hashing. In Proceedings of theAdvances in Neural Information Processing Systems, pages 1753–1760, 2008.
[8]Wei Liu, JunWang, Rongrong Ji, Yu-Gang Jiang, and Shih-Fu Chang. Supervised hashing withkernels. In Proceedings of the IEEE Conference on Computer Vision and PatternRecognition, pages 2074–2081, 2012.
[9]Wei Liu, Cun Mu,Sanjiv Kumar, and ShihFu Chang. Discrete graph hashing. In Proceedings of theAdvances in Neural Information Processing Systems, pages 3419–3427, 2014.
4.获取paper
看到此篇paper是发表在IJCAI上的,从图书馆的数据库中获取此篇paper。
5.浏览paper
浏览大致内容,对每篇paper进行一个大致的评估,确定第3步中选择的都是你需要的paper。
6.抽取每篇paper的特征并进行总结
对这几篇paper提到的哈希方法进行总结:
哈希分为数据独立和数据依赖的哈希,传统哈希如LSH是数据独立的,数据依赖的哈希从训练数据中得到哈希函数,精度更高编码更短[2][4][8]。存在的数据依赖哈希又可以分为有监督的和无监督的,无监督的方法包括谱哈希(SH)[7],锚点哈希(AGH)[6],离散图哈希(DGH)[9]等,而SGH在图哈希算法中需要的运行时间短且精度更高。