原文來源:http://blog.csdn.net/Ying_Xu/article/details/50532185
——————————————————————————————————————————————————————————————————————————————————————————————————————————————————————————————
這一篇主要對圖像哈希技術(shù)的現(xiàn)有算法做一個研究性的概述。由于哈希函數(shù)的本質(zhì)是一個降維的操作律歼,因此會存在特征信息的丟失和檢索準確率的下降。
目前的哈希函數(shù)可以分為數(shù)據(jù)依賴的(Data-dependent)和數(shù)據(jù)獨立的(Data-independent),數(shù)據(jù)依賴的哈希函數(shù)也就是依賴原始數(shù)據(jù)來學(xué)習(xí)哈希函數(shù)歪玲,數(shù)據(jù)獨立的就是人工指定哈希函數(shù)。
最最經(jīng)典的也算作最原始的用于圖像檢索的哈希算法是LSH,即局部敏感哈希(Locality Sensitivehashing)。它是由Piotr Indyk等人提出的鸣皂,該方法對數(shù)據(jù)進行隨機映射,屬于數(shù)據(jù)獨立的哈希函數(shù)暮蹂。容易實現(xiàn)寞缝,計算速度也較快。這是一種非數(shù)據(jù)驅(qū)動型的算法仰泻,檢索精度并不高荆陆。
之后的很多哈希函數(shù)都是基于該LSH方法做出的改進和擴展延伸,如Jianqiu Ji等人提出的超比特局部敏感哈希(Super-BitLocality-Sensitive Hashing, SBLSH)集侯,以角度作為核函數(shù)度量標準被啼,對隨機投影向量進行分組正交化;Brian Kulis等人提出的核化局部敏感哈希(KernelizedLocality-Sensitive Hashing, KLSH)棠枉,對LSH進行了擴展趟据,利用核函數(shù)和圖像庫中的稀疏集來構(gòu)造隨機映射,可以選擇任意核函數(shù)作為相似性度量函數(shù)术健。
對于除了數(shù)據(jù)本身所具有的信息,數(shù)據(jù)可能還具有附加的信息粘衬,例如標簽信息等荞估,在模式識別、計算機視覺和機器學(xué)習(xí)等領(lǐng)域有著非常重要的作用稚新。因此基于此勘伺,哈希函數(shù)還可以分為基于監(jiān)督的、基于半監(jiān)督的和基于非監(jiān)督的哈希函數(shù)褂删。
這里對近幾年來的哈希函數(shù)做了一些總結(jié)和概述飞醉,整理在一個文檔中。是根據(jù)袁勇學(xué)長的一篇博客進行整理的屯阀,貼出來也供大家一起學(xué)習(xí)缅帘。
總結(jié)下載鏈接:http://pan.baidu.com/s/1bobfuzl? ?提取密碼:b97a
或者在這里下載
http://download.csdn.net/detail/ying_xu/9408905
這個文檔是excel總結(jié)的,下載下來看起來會很清晰难衰,預(yù)覽的格式亂了钦无,看起來很亂。建議下載盖袭。
但是沒有涵蓋大部分2014和2015年的相關(guān)paper失暂。
這里根據(jù)袁勇學(xué)長的總結(jié)也一并貼出來彼宠。
CVPR14 圖像檢索papers——圖像檢索
1.??Triangulation embedding and democratic aggregation for imagesearch (Orals)
2.??Collaborative Hashing (post)
3.??Packing and Padding: Coupled Multi-index for Accurate ImageRetrieval (post)?technical report
4.??Bayes Merging of Multiple Vocabularies for Scalable ImageRetrieval (post)?technical report
5.??Fast Supervised Hashing with Decision Trees for High-DimensionalData (post)
6.??Learning Fine-grained Image Similarity with Deep Ranking (post)
7.??Congruency-Based Reranking (post)可能
8.??Fisher and VLAD with FLAIR (post)可能
9.??Locality in Generic Instance Search from One Example (post)
10.? Asymmetric sparse kernelapproximations for large-scale visual search (post)
11.? Locally Linear Hashing forExtracting Non-Linear Manifolds (post)
12.? Adaptive Object Retrievalwith Kernel Reconstructive Hashing (post)
13.? Hierarchical Feature Hashingfor Fast Dimensionality Reduction (post)
CVPR15image retrieval reading list
Image retrieval關(guān)鍵詞
·????????FAemb: A Function Approximation-Based Embedding Method for Image Retrieval
·????????Image Retrieval Using Scene Graphs
·????????Revisiting Kernelized Locality-Sensitive Hashing for Improved Large-ScaleImage Retrieval
·????????Early Burst Detection for Memory-Efficient Image Retrieval
·????????Deep Semantic Ranking Based Hashing for Multi-Label Image Retrieval(已讀)
·????????Query-Adaptive Late Fusion for Image Search and Person Re-identification
Hashing關(guān)鍵詞
·????????Supervised Discrete Hashing
·????????Hashing With Binary Autoencoders
·????????Reflectance Hashing for Material Recognition
·????????Deep Hashing for Compact Binary Codes Learning
·????????Online Sketching Hashing
·????????Semantics-Preserving Hashing for Cross-View Retrieval
·????????Face Video Retrieval With Image Query via Hashing Across Euclidean Spaceand Riemannian Manifold
2016
· ? ? Learning to Hash for Indexing?Big Data——A Survey
This paper provides readers with a systematic understanding of insights, pros, and cons?of the emerging indexing and search methods for Big Data.
By Jun Wang, Member IEEE, Wei Liu, Member IEEE, Sanjiv Kumar, Member IEEE, and?Shih-Fu Chang, Fellow IEEE?
要想對大數(shù)據(jù)哈希有一個清晰和透徹的了解,我非常推薦以上這篇2016年1月的文章弟塞,需要反復(fù)研讀凭峡。
其中包含一些大數(shù)據(jù)現(xiàn)狀與趨勢的剖析,是來自與李武軍老師2015年的一篇中文paper决记,鏈接都包含如下摧冀,建議都仔細研讀一下,對于該研究方向的同學(xué)會獲益匪淺霉涨。
參考資料:
1. ? Hashing圖像檢索源碼及數(shù)據(jù)庫總結(jié)http://yongyuan.name/blog/codes-of-hash-for-image-retrieval.html
2.??大數(shù)據(jù)哈希學(xué)習(xí):現(xiàn)狀與趨勢http://www.36dsj.com/archives/23799
3. ?CVPR14圖像檢索papershttp://yongyuan.name/blog/cvpr14-reading-list.html
4.??CVPR15 image retrieval readinglisthttp://yongyuan.name/blog/cvpr15-image-retrieval-reading-list.html
版權(quán)聲明:本文為博主原創(chuàng)文章按价,未經(jīng)博主允許不得轉(zhuǎn)載。