发布时间: 2021-11-16 摘要点击次数: 全文下载次数: DOI: 10.11834/jig.200212 2021 | Volume 26 | Number 11 图像分析和识别

1. 湖北工业大学计算机学院, 武汉 430068;
2. 武汉理工大学理学院, 武汉 430070
 收稿日期: 2020-06-17; 修回日期: 2020-11-20; 预印本日期: 2020-11-27 基金项目: 中央高校基本科研业务费专项资金资助(2019IB010)；湖北省教育厅指导性项目(B2019049) 作者简介: 万方, 男, 讲师, 硕士生导师, 主要研究方向为计算机视觉和图像检索。E-mail: 20021026@hbut.edu.cn 强浩鹏, 男, 硕士研究生, 主要研究方向为深度学习和跨模态检索。E-mail: qianghaopengshuxuejd@whut.edu.cn 雷光波, 女, 讲师, 主要研究方向为视觉SLAM和深度学习。E-mail: 64694786@qq.com *通信作者: 万方  20021026@hbut.edu.cn 中图法分类号: P23 文献标识码: A 文章编号: 1006-8961(2021)11-2659-11

关键词

Self-supervised deep discrete hashing for image retrieval
Wan Fang1, Qiang Haopeng2, Lei Guangbo1
1. School of Computing, Hubei University of Technology, Wuhan 430068, China;
2. College of Science, Wuhan University of Technology, Wuhan 430070, China
Supported by: Fundamental Research Funds for the Central University (2019IB010); Guiding Project of Hubei Provincial Department of Education (B2019049)

Abstract

Objective Hashing techniques have attracted much attention and are widely applied in the nearest neighbor search for image retrieval on large-scale datasets due to the low storage cost and fast retrieval speed. With the great development of deep learning, deep neural networks have been widely incorporated in image hashing retrieval, and existing deep learning-based hashing methods demonstrate the effectiveness of the end-to-end deep learning architecture for hashing learning. However, these methods have several problems. First, these existing deep hashing methods ignore the guiding role of deep image feature information in training deep hashing functions. Second, most deep hashing methods are to solve a relaxed problem first to simplify the optimization involved in a binary code learning procedure and then quantize the solved continuous solution to achieve the approximate binary solution. This optimization strategy leads to a large binary quantization error, result ing in the generation of suboptimal hash codes. Thus, to solve these two problems, a self-supervised deep discrete hashing method (SSDDH) is proposed in this study. Method The proposed SSDDH consists of two steps. First, using matrix decomposition, the binary hash code is obtained by solving the self-supervised loss function composed of the deep feature matrix extracted by the convolutional neural network and the image label matrix. The obtained binary hash code is used as the supervision information to guide the training of deep hash function. Second, a pair-wise loss function is constructed to maintain the similarity between the hash codes generated by deep hash function while maintaining the similarity between these hash codes and binary hash codes. The discrete optimization algorithm is used to solve the optimal solution of the objective function, thus effectively reducing the binary quantization error. Result Several experiments are conducted on three public datasets to validate the performance of the proposed algorithm. The first experiment compares the mean average precision (mAP) values of different existing hash methods on different hash code lengths, including unsupervised methods, supervised shallow methods, and supervised deep methods. The performance experimental results show that the mAP of our method SSDDH achieves the best performance in all cases with different values of the code length. On the CIFAR-10 and NUS-WIDE(web image dataset from National University of Singapore) datasets, the mAP of SSDDH is 3% higher than the next highest method named DPSH(deep pairwise-supervised hashing). On the Flickr dataset, SSDDH is also 1% higher than the highest method DPSH. The second experiment involves the CIFAR-10 dataset. The precision recall (PR) curves of DPSH and SSDDH are plotted. Query result comparison shows the PR curves of DPSH and SSDDH with 48-bit hash codes on CIFAR-10, and our SSDDH remarkably outperforms its competitor. SSDDH and DPSH are also compared in terms of the accuracy of the top 20 returned images when the hash code length is 48 bits. The result of the experiment is visualized for easy observation. We also found that the retrieval performance of SSDDH is considerably higher than that of DPSH. Experiment 3 is designed for parameter sensitivity analysis of SSDDH. Here, a parameter is used, while the others are fixed. Our method is insensitive to the parameters. This finding relatively demonstrates the robustness and effectiveness of the proposed method. Experiment 4 is conducted on CIFAR-10 when the hash code length is 48 bits to explore the difference between DPSH and SSDDH in time complexity. At the later stage of model training, SSDDH performance is better than DPSH at the same time consumption. Conclusion Considering that the existing deep hash methods ignore the guiding role of deep image feature information in the training of deep hash function and have the problem of large binary quantization error, this study proposes a self-supervised deep discrete hashing method named SSDDH. The deep feature matrix extracted by the convolutional neural network and the image label matrix are used to obtain the binary hash codes and make the binary hash codes the supervised information to guide the training of deep hash function. The similarity between the hash codes generated by deep hash function and the similarity between these hash codes and binary hash codes are maintained by constructing a pair-wise loss function. The binary quantization error is effectively reduced using the discrete cyclic coordinate descent. Comparison with several existing methods on three commonly used public datasets proves that this method is more efficient than the existing hash retrieval method. Future work lies in two aspects: First, focus will be on learning better fine-grained representation with more effectively. Second, semi-supervised regularization will be applied to our framework to make full use of the unlabeled data. Both will be employed to boost the image retrieval accuracy further. Third, our current approach will be extended to cross-modal retrieval, such as given a text query, to obtain all semantic relevant images from the database.

Key words

deep learning; image retrieval; hash learning; self-supervised; discrete optimization

1 深度成对监督哈希

 $p\left(s_{i j} \mid \boldsymbol{B}\right)= \begin{cases}\sigma\left(\varOmega_{i j}\right) & s_{i j}=1 \\ 1-\sigma\left(\varOmega_{i j}\right) & s_{i j}=0\end{cases}$ (1)

 \begin{aligned} \min \limits_{\boldsymbol{B}} J_{1}=&-\log p(\boldsymbol{S} \mid \boldsymbol{B})=-\sum\limits_{s_{i j} \in \boldsymbol{S}} \log p\left(s_{i j} \mid \boldsymbol{B}\right)=\\ &-\sum\limits_{s_{i j} \in \boldsymbol{S}}\left(s_{i j} {\varOmega}_{i j}-\log \left(1+\mathrm{e}^{\varOmega_{i j}}\right)\right) \end{aligned} (2)

 $\begin{gathered} \min \limits_{\boldsymbol{B}, \boldsymbol{H}} J_{2}=-\sum\limits_{s_{i j} \in \boldsymbol{S}}\left(s_{i j} \boldsymbol{\varTheta}_{i j}-\log \left(1+\mathrm{e}^{\boldsymbol{\varTheta}_{i j}}\right)\right)+ \\ \eta \sum\limits_{i=1}^{n}\left\|\boldsymbol{b}_{i}-\boldsymbol{h}_{i}\right\|_{\mathrm{F}}^{2} \end{gathered}$ (3)

 $\begin{gathered} \min \limits_{\boldsymbol{B}, \boldsymbol{H}} J_{3}=-\sum\limits_{s_{i j} \in \boldsymbol{S}}\left(s_{i j} \boldsymbol{\varTheta}_{i j}-\log \left(1+\mathrm{e}^{\boldsymbol{\varTheta}_{i j}}\right)\right)+ \\ \mu\left\|\boldsymbol{L}-\boldsymbol{B} \boldsymbol{W}^{\mathrm{T}}\right\|_{\mathrm{F}}^{2}+v\|\boldsymbol{W}\|_{\mathrm{F}}^{2} \end{gathered}$ (4)

 $\begin{gathered} \min \limits_{\boldsymbol{B}, \boldsymbol{H}} J_{5}=-\sum\limits_{s_{i j} \in \boldsymbol{S}}\left(s_{i j} \boldsymbol{\varTheta}_{i j}-\log \left(1+\mathrm{e}^{\boldsymbol{\varTheta}_{i j}}\right)\right)+ \\ \sum\limits_{s_{i j} \in \boldsymbol{S}^{\prime}}\left(\boldsymbol{h}_{i} \boldsymbol{b}_{j}^{\mathrm{T}}-k s_{i j}^{\prime}\right)^{2}+\alpha\|\boldsymbol{B}-\boldsymbol{H}\|_{\mathrm{F}}^{2}+ \\ \beta\|\boldsymbol{H} {\bf{1}}\|_{\mathrm{F}}^{2}+\gamma\|\hat{\boldsymbol{L}}-\boldsymbol{L}\|_{\mathrm{F}}^{2} \\ \text { s. t. } \quad \boldsymbol{B} \in\{-1,+1\}^{n \times k} \\ \qquad\ \ \ \boldsymbol{H} \in[-1,+1]^{n \times k} \end{gathered}$ (6)

2.3 目标损失函数以及算法求解

 $\begin{gathered} \min \limits_{\theta, \boldsymbol{B}, \boldsymbol{U}, \boldsymbol{W}} J=J_{4}+J_{5}= \\ \left\|\boldsymbol{F}-\boldsymbol{B} \boldsymbol{U}^{\mathrm{T}}\right\|_{\mathrm{F}}^{2}+\left\|\boldsymbol{L}-\boldsymbol{B} \boldsymbol{W}^{\mathrm{T}}\right\|_{\mathrm{F}}^{2}- \\ \sum\limits_{s_{i j} \in \boldsymbol{S}}\left(s_{i j} \boldsymbol{\varTheta}_{i j}-\log \left(1+\mathrm{e}^{\boldsymbol{\varTheta}_{i j}}\right)\right)+ \\ \left\|\boldsymbol{H} \boldsymbol{B}^{\mathrm{T}}-k \boldsymbol{S}^{\prime}\right\|_{\mathrm{F}}^{2}+\alpha\|\boldsymbol{B}-\boldsymbol{H}\|_{\mathrm{F}}^{2}+ \\ \beta\|\boldsymbol{H} {\bf{1}}\|_{\mathrm{F}}^{2}+\gamma\|\hat{\boldsymbol{L}}-\boldsymbol{L}\|_{\mathrm{F}}^{2} \\ \text { s. t. } \quad \boldsymbol{B} \in\{-1,+1\}^{n \times k} \end{gathered}$ (7)

1) 更新参数$\theta$，固定其他参数的值不变，则式(7)可以改写为

 $\begin{gathered} \min _{\theta} J=-\sum\limits_{s_{i j} \in \boldsymbol{S}}\left(s_{i j} \boldsymbol{\varTheta}_{i j}-\log \left(1+\mathrm{e}^{\boldsymbol{\varTheta}_{i j}}\right)\right)+ \\ \left\|\boldsymbol{H B}^{\mathrm{T}}-k \boldsymbol{S}^{\prime}\right\|_{\mathrm{F}}^{2}+\alpha\|\boldsymbol{B}-\boldsymbol{H}\|_{\mathrm{F}}^{2}+ \\ \beta\|\boldsymbol{H} {\bf{1}}\|_{\mathrm{F}}^{2}+\gamma\|\hat{\boldsymbol{L}}-\boldsymbol{L}\|_{\mathrm{F}}^{2} \end{gathered}$ (8)

2) 更新参数$\mathit{\boldsymbol{U}}$，固定其他参数的值不变，则式(7)可以改写为

 $\min \limits_{\boldsymbol{U}} \boldsymbol{J}=\left\|\boldsymbol{F}-\boldsymbol{B} \boldsymbol{U}^{\mathrm{T}}\right\|_{\mathrm{F}}^{2}$ (9)

 $\boldsymbol{U}=\boldsymbol{F}^{\mathrm{T}} \boldsymbol{B}\left(\boldsymbol{B}^{\mathrm{T}} \boldsymbol{B}\right)^{-1}$ (10)

3) 更新参数$\mathit{\boldsymbol{W}}$，固定其他参数的值不变，则式(7)可以改写为

 $\min \limits_{\boldsymbol{W}} \boldsymbol{J}=\left\|\boldsymbol{L}-\boldsymbol{B} \boldsymbol{W}^{\mathrm{T}}\right\|_{\mathrm{F}}^{2}$ (11)

 $\boldsymbol{W}=\boldsymbol{L}^{\mathrm{T}} \boldsymbol{B}\left(\boldsymbol{B}^{\mathrm{T}} \boldsymbol{B}\right)^{-1}$ (12)

4) 更新参数$\mathit{\boldsymbol{B}}$，固定$\theta $$\mathit{\boldsymbol{U}}$$\mathit{\boldsymbol{W}}$的值，则式(7)可以改写为

 $\begin{gathered} \min \limits_{\boldsymbol{B}} J=\left\|\boldsymbol{F}-\boldsymbol{B} \boldsymbol{U}^{\mathrm{T}}\right\|_{\mathrm{F}}^{2}+\left\|\boldsymbol{L}-\boldsymbol{B} \boldsymbol{W}^{\mathrm{T}}\right\|_{\mathrm{F}}^{2}+ \\ \left\|\boldsymbol{H} \boldsymbol{B}^{\mathrm{T}}-k \boldsymbol{S}^{\prime}\right\|_{\mathrm{F}}^{2}+\alpha\|\boldsymbol{B}-\boldsymbol{H}\|_{\mathrm{F}}^{2} \\ \text { s. t. } \quad \boldsymbol{B} \in\{-1,+1\}^{n \times k} \end{gathered}$ (13)

 $\left\|\boldsymbol{F}-\boldsymbol{B} \boldsymbol{U}^{\mathrm{T}}\right\|_{\mathrm{F}}^{2}=\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{U}^{\mathrm{T}} \boldsymbol{U} \boldsymbol{B}^{\mathrm{T}}\right)+\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{Q}_{1}^{\mathrm{T}}\right)+c$

 $\left\|\boldsymbol{L}-\boldsymbol{B} \boldsymbol{W}^{\mathrm{T}}\right\|_{\mathrm{F}}^{2}=\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{W}^{\mathrm{T}} \boldsymbol{W} \boldsymbol{B}^{\mathrm{T}}\right)+\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{Q}_{2}^{\mathrm{T}}\right)+c$

 $\left\|\boldsymbol{H} \boldsymbol{B}^{\mathrm{T}}-k \boldsymbol{S}^{\prime}\right\|_{\mathrm{F}}^{2}=\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{H}^{\mathrm{T}} \boldsymbol{H} \boldsymbol{B}^{\mathrm{T}}\right)+\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{Q}_{3}^{\mathrm{T}}\right)+c$

 $\boldsymbol{\alpha}\|\boldsymbol{B}-\boldsymbol{H}\|_{\mathrm{F}}^{2}=\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{Q}_{4}^{\mathrm{T}}\right)+c$

 $\begin{gathered} \min \limits_{\boldsymbol{B}} J= \operatorname{tr}\left(\boldsymbol{B} \boldsymbol{U}^{\mathrm{T}} \boldsymbol{U} \boldsymbol{B}^{\mathrm{T}}\right)+\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{W}^{\mathrm{T}} \boldsymbol{W} \boldsymbol{B}^{\mathrm{T}}\right)+\\ \operatorname{tr}\left(\boldsymbol{B} \boldsymbol{Q}_{1}\right)+\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{Q}_{2}\right)+\\ \operatorname{tr}\left(\boldsymbol{B} \boldsymbol{Q}_{3}\right)+\operatorname{tr}\left(\boldsymbol{B} \boldsymbol{Q}_{4}\right)+c \\ \text { s. t. } \quad \boldsymbol{B} \in\{-1,+1\}^{n \times k} \end{gathered}$ (14)

 $\begin{gathered} \min \limits_{\boldsymbol{B}_{*{r}}} J\left(\boldsymbol{B}_{*{r}}\right)= \operatorname{tr}\left(\boldsymbol { B } _ { * { r } } \left[2 \boldsymbol{U}_{*{r}}^{\mathrm{T}} \hat{\boldsymbol{U}}_{r} \hat{\boldsymbol{B}}_{r}^{\mathrm{T}}+2 \boldsymbol{W}_{*{r}}^{\mathrm{T}} \hat{\boldsymbol{W}}_{r} \hat{\boldsymbol{B}}_{r}^{\mathrm{T}}+\right.\right.\\ 2 \boldsymbol{H}_{*{r}}^{\mathrm{T}} \hat{\boldsymbol{H}}_{r} \hat{\boldsymbol{B}}_{r}^{\mathrm{T}}+\left(\boldsymbol{Q}_{1}\right)_{*{r}}^{\mathrm{T}}+\left(\boldsymbol{Q}_{2}\right)_{*{r}}^{\mathrm{T}}+ \\ \left.\left.\left(\boldsymbol{Q}_{3}\right)_{*{r}}^{\mathrm{T}}+\left(\boldsymbol{Q}_{4}\right)_{*{r}}^{\mathrm{T}}\right]\right) \\ \text { s.t. } \quad \boldsymbol{B}_{*{r}} \in\{-1,+1\}^{n} \end{gathered}$ (15)

 \begin{aligned} \boldsymbol{B}_{*{r}} &=-\operatorname{sign}\left(2 \hat{\boldsymbol{B}}_{r}\left[\hat{\boldsymbol{U}}_{r}^{\mathrm{T}} \boldsymbol{U}_{*{r}}+\hat{\boldsymbol{W}}_{r}^{\mathrm{T}} \boldsymbol{W}_{*{r}}+\hat{\boldsymbol{H}}_{r}^{\mathrm{T}} \boldsymbol{H}_{*{r}}\right]+\right.\\ &\left.\left(\boldsymbol{Q}_{1}\right)_{*{r}}+\left(\boldsymbol{Q}_{2}\right)_{*{r}}+\left(\boldsymbol{Q}_{3}\right)_{*{r}}+\left(\boldsymbol{Q}_{4}\right)_{*{r}}\right) \end{aligned} (16)

for iter = 1，2，…，$T$ do;

for i = 1，2，…，$t$ do;

end for;

for $r$ =1→$k$ do;

end for;

end for。

2.4 样本外扩展

 $\boldsymbol{b}_{q}=\operatorname{sign}\left(\boldsymbol{F}\left(\boldsymbol{x}_{q} ; \theta\right)\right)$ (17)

3.1 数据集与评价指标

CIFAR-10包含60 000幅32×32像素的彩色图像，共10类，每幅图像属于其中的一类。NUS-WIDE包含近270 000幅彩色图像，共81类，每幅图像至少属于其中一类，遵循Lai等人(2015)的实验设置，实验只使用其中与最常见的21类相关的图像数据。Flickr包含从Flickr网站搜集的20 015幅图像，共24类，每幅图像至少属于其中一类语义标签。

 $A P(\boldsymbol{q})=\frac{1}{Q} \sum\limits_{r=1}^{N} P_{q}(r) {rel}(r)$ (18)

3.3 对比实验与分析

Table 1 mAPs results of different hashing methods on three datasets

 方法 CIFAR-10 NUS-WIDE Flickr 12 bit 24 bit 32 bit 48 bit 12 bit 24 bit 32 bit 48 bit 12 bit 24 bit 32 bit 48 bit DPSH (Li等，2016) 0.713 0.727 0.744 0.757 0.752 0.790 0.794 0.812 0.837 0.844 0.856 0.863 HashGAN (Cao等，2018) 0.660 0.722 0.731 0.735 0.642 0.699 0.737 0.751 0.790 0.826 0.841 0.848 DHN (Zhu等，2016) 0.555 0.594 0.603 0.621 0.708 0.735 0.748 0.758 0.823 0.839 0.844 0.847 DSH (Jin等，2014) 0.616 0.651 0.66 0.675 0.548 0.551 0.558 0.562 0.805 0.835 0.842 0.845 SDH (Shen等，2015) 0.285 0.329 0.341 0.356 0.568 0.600 0.608 0.637 Nan Nan Nan Nan ITQ (Gong等，2013) 0.162 0.169 0.172 0.175 0.452 0.468 0.472 0.477 0.684 0.695 0.697 0.697 SH (Weiss等，2008) 0.127 0.128 0.126 0.129 0.454 0.406 0.405 0.400 0.645 0.651 0.65 0.646 本文 0.749 0.762 0.774 0.785 0.787 0.811 0.816 0.823 0.840 0.852 0.862 0.871 注：加粗字体为各列最优结果, Nan表示该方法在数据集上没有实验数据。

参考文献

• Andoni A and Indyk P. 2006. Near-optimal hashing algorithms for approximate nearest neighbor in high dimensions//The 47th Annual IEEE Symposium on Foundations of Computer Science. Berkeley, UK: IEEE: 459-468[DOI: 10.1109/FOCS.2006.49]
• Cao Y, Liu B, Long M S and Wang J M. 2018. HashGAN: deep learning to hash with pair conditional Wasserstein GAN//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE: 1287-1296[DOI: 10.1109/CVPR.2018.00140]
• Chatfield K, Simonyan K, Vedaldi A and Zisserman A. 2014. Return of the devil in the details: delving deep into convolutional nets//Proceedings of the British Machine Vision Conference. Nottingham, UK: BMVA Press: 1-12[DOI: 10.5244/C.28.6]
• Chua T S, Tang J H, Hong R C, Li H J, Luo Z P and Zheng Y T. 2009. NUS-WIDE: a real-world web image database from national university of Singapore//Proceedings of the ACM International Conference on Image and Video Retrieval. Santorini, Greece: ACM: 1-9[DOI: 10.1145/1646396.1646452]
• Deng J, Dong W, Socher R, Li L J, Li K and Li F F. 2010. ImageNet: a large-scale hierarchical image database//Proceedings of 2009 IEEE Conference on Computer Vision and Pattern Recognition. Miami, USA: IEEE: 248-255[DOI: 10.1109/cvpr.2009.5206848]
• Gong Y C, Lazebnik S, Gordo A, Perronnin F. 2013. Iterative quantization: a procrustean approach to learning binary codes for large-scale image retrieval. IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(12): 2916-2929 [DOI:10.1109/TPAMI.2012.193]
• Goodfellow I J, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley D, Ozair S, Courville A and Bengio Y. 2014. Generative adversarial nets//Proceedings of the 27th International Conference on Neural Information Processing Systems. Montreal, Canada: MIT Press: 2672-2680
• He K M, Gkioxari G, Dollár P, Girshick R. 2020. Mask R-CNN. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(2): 386-397 [DOI:10.1109/TPAMI.2018.2844175]
• Huiskes M J and Lew M S. 2008. The MIR Flickr retrieval evaluation//Proceedings of the 1st ACM International Conference on Multimedia Information Retrieval. Vancouver, Canada: ACM: 39-43[DOI: 10.1145/1460096]
• Jiang Q Y and Li W J. 2017. Deep cross-modal hashing//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE: 3270-3278[DOI: 10.1109/CVPR.2017.348]
• Jiang Q Y, Cui X, Li W J. 2018. Deep discrete supervised hashing. IEEE Transactions on Image Processing, 27(12): 5996-6009 [DOI:10.1109/TIP.2018.2864894]
• Jin Z M, Li C, Lin Y, Cai D. 2014. Density sensitive hashing. IEEE Transactions on Cybernetics, 44(8): 1362-1371 [DOI:10.1109/TCYB.2013.2283497]
• Krizhevsky A. 2009. Learning Multiple Layers of Features from Tiny Images. Technical Report TR-2009. University of Toronto
• Krizhevsky A, Sutskever I and Hinton G E. 2012. ImageNet classification with deep convolutional neural networks//Proceedings of the 26th International Conference on Neural Information Processing Systems. Montreal, Canada: MIT Press: 1097-1105
• Kulis B and Grauman K. 2009. Kernelized locality-sensitive hashing for scalable image search//Proceedings of the 12th IEEE International Conference on Computer Vision. Kyoto, Japan: IEEE: 2130-2137[DOI: 10.1109/ICCV.2009.5459466]
• Lai H J, Pan Y, Liu Y and Yan S C. 2015. Simultaneous feature learning and hash coding with deep neural networks//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston, USA: IEEE: 3270-3278[DOI: 10.1109/CVPR.2015.7298947]
• Li Q, Sun Z N, He R and Tan T N. 2017. Deep supervised discrete hashing//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, USA: Curran Associates Inc. : 2482-2491
• Li W J, Wang S and Kang W C. 2016. Feature learning based deep supervised hashing with pairwise labels//Proceedings of the 25th International Joint Conference on Artificial Intelligence. New York, USA: AAAI: 1711-1717
• Liu H M, Wang R P, Shan S G and Chen X L. 2016a. Deep supervised hashing for fast image retrieval//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE: 2064-2072[DOI: 10.1109/CVPR.2016.227]
• Liu W, Anguelov D, Erhan D, Szegedy C, Reed S, Fu C Y and Berg A C. 2016b. SSD: single shot MultiBox detector//Proceedings of the 14th European Conference on Computer Vision. Amsterdam, the Netherlands: Springer: 398-413[DOI: 10.1007/978-3-319-46448-0_2]
• Liu W, Mu C, Kumar S and Chang S F. 2014. Discrete graph hashing//Proceedings of the 27th International Conference on Neural Information Processing Systems. Montreal, Canada: MIT Press: 3419-3427
• Liu W, Wang J, Ji R R, Jiang Y G and Chang S F. 2012. Supervised hashing with kernels//Proceedings of 2012 IEEE Conference on Computer Vision and Pattern Recognition. Providence, USA: IEEE: 2074-2081[DOI: 10.1109/CVPR.2012.6247912]
• Liu Y, Pan Y, Xia R K, Liu D, Yin J. 2016. FP-CNN: a fast image hashing algorithm based on deep convolutional neural network. Computer Science, 43(9): 39-46, 51 (刘冶, 潘炎, 夏榕楷, 刘荻, 印鉴. 2016. FP-CNNH: 一种基于深度卷积神经网络的快速图像哈希算法. 计算机科学, 43(9): 39-46, 51) [DOI:10.11896/j.issn.1002-137X.2016.9.007]
• Lu X Q, Chen Y X, Li X L. 2020. Siamese dilated inception hashing with intra-group correlation enhancement for image retrieval. IEEE Transactions on Neural Networks and Learning Systems, 31(8): 3032-3046 [DOI:10.1109/TNNLS.2019.2935118]
• Noh H, Hong S and Han B. 2015. Learning deconvolution network for semantic segmentation//Proceedings of 2015 IEEE International Conference on Computer Vision. Santiago, Chile: IEEE: 1520-1528[DOI: 10.1109/ICCV.2015.178]
• Ren S Q, He K M, Girshick R, Sun J. 2017. Faster R-CNN: towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(6): 1137-1149 [DOI:10.1109/TPAMI.2016.2577031]
• Shen F M, Shen C H, Liu W and Shen H T. 2015. Supervised discrete hashing//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston, USA: IEEE: 37-45[DOI: 10.1109/CVPR.2015.7298598]
• Shen Y M, Feng Y, Fang B, Zhou M L, Kwong S, Qiang B H. 2020. DSRPH: deep semantic-aware ranking preserving hashing for efficient multi-label image retrieval. Information Sciences, 539: 145-156 [DOI:10.1016/j.ins.2020.05.114]
• Simonyan K and Zisserman A. 2015. Very deep convolutional networks for large-scale image recognition[EB/OL]. [2020-05-17]. http://arxiv.org/pdf/1409.1556.pdf
• Wang J, Kumar S and Chang S F. 2010. Sequential projection learning for hashing with compact codes//Proceedings of the 27th International Conference on International Conference on Machine Learning. Haifa, Israel: Omnipress: 1127-1134
• Wang Z M, Zhang H. 2019. A fast image retrieval method based on multi-layer CNN features. Journal of Computer-Aided Design and Computer Graphics, 31(8): 1410-1416 (王志明, 张航. 2019. 融合多层卷积神经网络特征的快速图像检索方法. 计算机辅助设计与图形学学报, 31(8): 1410-1416) [DOI:10.3724/SP.J.1089.2019.17845]
• Weiss Y, Torralba A and Fergus R. 2008. Spectral hashing//Proceedings of the 21st International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc. : 1753-1760
• Xia R K, Pan Y, Lai H J, Liu C and Yan S C. 2014. Supervised hashing for image retrieval via image representation learning//Proceedings of the 28th AAAI Conference on Artificial Intelligence. Québec City, Canada: AAAI: 2156-2162
• Yan C G, Xie H T, Yang D B, Yin J, Zhang Y D, Dai Q H. 2018. Supervised hash coding with deep neural network for environment perception of intelligent vehicles. IEEE Transactions on Intelligent Transportation Systems, 19(1): 284-295 [DOI:10.1109/TITS.2017.2749965]
• Zhang P C, Zhang W, Li W J and Guo M Y. 2014. Supervised hashing with latent factor models//Proceedings of the 37th International ACM SIGIR Conference on Research and Development in Information Retrieval. Gold Coast, Australia: ACM: 173-182[DOI: 10.1145/2600428.2609600]
• Zhu H, Long M S, Wang J M and Cao Y. 2016. Deep hashing network for efficient similarity retrieval//Proceedings of the 30th AAAI Conference on Artificial Intelligence. Phoenix, USA: AAAI: 2415-2421