论文笔记（关于图像检索的总结性论文）：Content-Based Image Retrieval and Feature Extraction: A Comprehensive Review（中）

继上篇：https://blog.csdn.net/timcanby/article/details/104382103 

今天继续对：

Section 6 底层特征的融合运用

Section 7 局部特征

Section 8 基于深度学习的种种

==============================进行整理===================================================

Section 6 底层特征的融合运用

[55] R. Ashraf, M. Ahmed, S. Jabbar et al., “Content based image retrieval by using color descriptor and discrete wavelet transform,” Journal of Medical Systems, vol. 42, no. 3, p. 44,2018.

这个的模型基于discrete wavelet transform (DWT)和颜色，使用了颜色（RGB 和 YCbCr）纹理和形状特征（Canny 边缘算子），检索用的ANN

[56] R. Ashraf, M. Ahmed, U. Ahmad, M. A. Habib, S. Jabbar, and K. Naseer, “MDCBIR-MF: multimedia data for content based image retrieval by using multiple features,” Multimedia Tools and Applications, pp. 1–27, 2018.

这个用了颜色和纹理特征，这篇处理颜色信息的color moment有点意思

[57] Y. Mistry, D. Ingole, and M. Ingole, “Content based image retrieval using hybrid features and various distance metric,” Journal of Electrical Systems and Information Technology,vol. 5, no. 3, pp. 878–888, 2017.

这篇总结了很多距离计算的方法的区别，可以看看

[58] K. T. Ahmed, M. A. Iqbal, and A. Iqbal, “Content based image retrieval using image features information fusion,”Information Fusion, vol. 51, pp. 76–99, 2018.

这篇文章用了颜色特征和边缘特征，然后用BoVW来检索

[59] P. Liu, J.-M. Guo, K. Chamnongthai, and H. Prasetyo,“Fusion of color histogram and LBP-based features for texture image retrieval and classification,” Information Sciences,vol. 390, pp. 95–111, 2017.

这个用了LBP（local base pattern），和颜色特征

[60] W. Zhou, H. Li, J. Sun, and Q. Tian, “Collaborative index embedding for image retrieval,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 40, no. 5,pp. 1154–1166, 2018.

这篇用了来自cnn的中间层输出的深度特征和SIFT特征进行Collaborative Index Embedding然后检索

[61] C. Li, Y. Huang, and L. Zhu, “Color texture image retrieval based on Gaussian copula models of Gabor wavelets,” Pattern Recognition, vol. 64, pp. 118–129, 2017.

如题这个用了Gabor wavelet

[62] H. H. Bu, N. Kim, C. J. Moon, and J. H. Kim, “Content-based image retrieval using combined color and texture features extracted by multi-resolution multi-direction filtering,”Journal of Information Processing Systems, vol. 13, no. 3, pp. 464–475, 2017.

这篇是用Multi-Resolution Multi-Directional (MRMD) filters来结合了颜色和纹理特征

[63] A. Nazir, R. Ashraf, T. Hamdani, and N. Ali, “Content based image retrieval system by using HSV color histogram,discrete wavelet transform and edge histogram descriptor,” in Proceedings of the 2018 International Conference on Computing, Mathematics and Engineering Technologies (iCoMET), pp. 1–6, IEEE, Sukkur, Pakistan,March 2018.

这个就是如题所示，用了HSV color histogram,discrete wavelet transform and edge histogram descriptor做了很多对比实验

然后Section 7 基于局部特征的检索

这个比较长，慢慢来

讲这个之前需要先讲一下接下来论文很多都会用到的：稀疏表示（sparse representation），这边参考一下这篇博客：

https://blog.csdn.net/Forever_pupils/article/details/88572281

还有这篇论文：

C. Celik and H. S. Bilge, “Content based image retrieval with sparse representations and local feature descriptors: a comparative study,” Pattern Recognition, vol. 68, pp. 1–13,2017.

https://www.sciencedirect.com/science/article/pii/S0031320317301048

[64] L.-W. Kang, C.-Y. Hsu, H.-W. Chen, C.-S. Lu, C.-Y. Lin, andS.-C. Pei, “Feature-based sparse representation for image similarity assessment,” IEEE Transactions on Multimedia,vol. 13, no. 5, pp. 1019–1030, 2011.

这篇里面有个比较关键的点叫：K-SVD dictionary learning algorithm 来自于：M. Aharon, M. Elad, and A. M. Bruckstein, “The K-SVD: An algorithm for designing of overcomplete dictionaries for sparse representation,” IEEE Trans. Signal Process., vol. 54, no. 11, pp. 4311–4322, Nov. 2006

[65] Z.-Q. Zhao, H. Glotin, Z. Xie, J. Gao, and X. Wu, “Cooperative sparse representation in two opposite directions for semi-supervised image annotation,” IEEE Transactions on Image Processing, vol. 21, no. 9, pp. 4218–4231, 2012.

这篇的主要就是自监督的一个模型Co-KSR的提出（然后本文介绍了一些有名的自监督分类器比如TSVM, GFHF,  LGC）

[66] J. J. )iagarajan, K. N. Ramamurthy, P. Sattigeri, and A. Spanias, “Supervised local sparse coding of sub-image features for image retrieval,” in Proceedings of the 2012 19th IEEE International Conference on Image Processing (ICIP),pp. 3117–3120, IEEE, Melbourne, Australia, September-October 2012.

这篇是supervised的

[67] C. Hong and J. Zhu, “Hypergraph-based multi-example ranking with sparse representation for transductive learning image retrieval,” Neurocomputing, vol. 101, pp. 94–103, 2013.

（https://reader.elsevier.com/reader/sd/pii/S0925231212006443?token=9E90AD2CBD6B406E1CB15A16D7A9CC86ABF6C996BF6C9FA52777AE88BF4C21E7ACE9081BBB19E89641C1030417291FAA）这篇推荐一下主要是解决了多图同时处理节约了计算时间的问题而且举例和表达都相对易懂。

[68] D. Wang, S. C. Hoi, Y. He, J. Zhu, T. Mei, and J. Luo,“Retrieval-based face annotation by weak label regularized local coordinate coding,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 36, no. 3, pp. 550–563,2014

这个是人脸分类上的运用

[69] M. Srinivas, R. R. Naidu, C. S. Sastry, and C. K. Mohan,“Content based medical image retrieval using dictionary learning,” Neuro computing, vol. 168, pp. 880–895, 2015.

K-SVD拿来聚类词典

[70] S. Mohamadzadeh and H. Farsi, “Content-based image retrieval system via sparse representation,” IET Computer

Vision, vol. 10, no. 1, pp. 95–102, 2016.

这篇好处是对比了很多现有的方法

[71] Q. Li, Y. Han, and J. Dang, “Sketch4Image: a novel framework for sketch-based image retrieval based on product quantization with coding residuals,” Multimedia Tools and Applications, vol. 75, no. 5, pp. 2419–2434, 2016.

这篇很好玩，基于sketch图像的，重视手上data形状轮廓特征的同学有很大参考价值

[72] H. Wu, R. Bie, J. Guo, X. Meng, and S. Wang, “Sparse coding based few learning instances for image retrieval,” Multimedia Tools and Applications, vol. 78, no. 5, pp. 6033–6047, 2018.

这篇组合了cross-validation sparse coding representation, sparse coding-based instance distance, 和 improved KNN model

[73] Y. Duan, J. Lu, J. Feng, and J. Zhou, “Context-aware local binary feature learning for face recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 40,no. 5, pp. 1139–1153, 2018.

这篇的特色在于取得了robust local binary features的提案，也就是对Binary Feature Descriptor的讨论，非常的推荐阅读一下，地址：https://ieeexplore.ieee.org/document/7936534

然后来个评价试验的结果：

然后作者总结了下基于ML经常用于CBIR的方法（如下图）：

接下来是很多同学最喜爱的，基于深度学习：

Section 8 基于深度学习的种种

[82] N. Kondylidis, M. Tzelepi, and A. Tefas, “Exploiting tf-idf in deep convolutional neural networks for content based image retrieval,” Multimedia Tools and Applications, vol. 77, no. 23, pp. 30729–30748, 2018.

这个把操作文本的tf-idf嵌入了CNN框架里使用的基础框架是VGG16

[83] X. Shi, M. Sapkota, F. Xing, F. Liu, L. Cui, and L. Yang,“Pairwise based deep ranking hashing for histopathology image classification and retrieval,” Pattern Recognition,vol. 81, pp. 14–22, 2018.

这篇作者提出了 deep ranking hashing来对医学图像进行01编码

[84] L. Zhu, J. Shen, L. Xie, and Z. Cheng, “Unsupervised visual hashing with semantic assistant for content-based image retrieval,” IEEE Transactions on Knowledge and Data Engineering,vol. 29, no. 2, pp. 472–486, 2017.

https://ink.library.smu.edu.sg/cgi/viewcontent.cgi?article=4813&context=sis_research

offline learning and online learning 的 Unsupervised学习，这篇文章强烈的标小星星啊，结合了图像的wiki文本描述，把语义和本体内容融合起来提出了一种编码的方法

[85] A. Alzu’bi, A. Amira, and N. Ramzan, “Content-based image retrieval with compact deep convolutional features,” Neurocomputing,vol. 249, pp. 95–105, 2017.

然后这个使用了深度卷积特征的

[89] A. Karpathy and L. Fei-Fei, “Deep visual-semantic alignments for generating image descriptions,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,pp. 3128–3137, Boston, MA, USA, June 2015.

然后其实检索的一大挑战就是处理具有很多隐藏语义的复杂图像，这是李飞飞的经典之作之一了，生成图像描述的（https://cs.stanford.edu/people/karpathy/cvpr2015.pdf）

[92] C. Zhang, J. Cheng, and Q. Tian, “Multiview, few-labeled object categorization by predicting labels with view consistency,”IEEE Transactions

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8398481

然后基于深度学习基础的我觉得这篇文章写的太跳跃了之后新开一篇来写

 

剩余的将会在下一篇博客（下）里更新，重点讲解

本次两块的内容，如果有什么错误欢迎指正留言～～

=======================================================

个人github：https://github.com/timcanby