﻿ 基于复杂网络的图像特征提取及多特征融合方案探究 Image Feature Extraction Based on Complex Network and Multi-Feature Fusion Schemes Exploration in CBIR

Journal of Image and Signal Processing
Vol.04 No.04(2015), Article ID:16197,10 pages
10.12677/JISP.2015.44012

Image Feature Extraction Based on Complex Network and Multi-Feature Fusion Schemes Exploration in CBIR

Jibin Gao1,2, Yumei Li1*, Huina Zhang1

1Department of Mathematics, School of Science, Beijing Technology and Business University, Beijing

2College of Science, Guilin University of Technology, Guilin Guangxi

Email: *liwjyumei@163.com

Received: Oct. 3rd, 2015; accepted: Oct. 16th, 2015; published: Oct. 21st, 2015

ABSTRACT

Image shape feature’s extraction is an important research content in content-based image retrieval, and an image shape feature extraction method by using complex network model is proposed in this paper. First, SIFT keypoints of an image are extracted, and then the image is divided into blocks such that the initial complex network model can be built in each block respectively. After that, minimum span ning tree decomposition method is used for the network’s dynamic evolution, and the network features at different moments in different blocks are extracted as the image’s shape features. Furthermore, the shape features are combined with the color and texture features and a kind of fusion feature is obtained. By experiment results comparison, it shows that the fusion feature does have advantages in CBIR.

Keywords:Content-Based Image Retrieval, Complex Network, SIFT Keypoints, Topological Structure, Multi-Feature Fusion

1北京工商大学理学院数学系，北京

2桂林理工大学理学院，广西 桂林

Email: *liwjyumei@163.com

1. 概述

2. 基于复杂网络的图像特征提取

2.1. 特征描述与动态演化

2.2. 利用图像关键点构建复杂网络

2.3. 复杂网络的动态演化实现

2.4. 基于复杂网络拓扑结构的特征提取

Figure 1. Keypoints partition and its dynamic evolution of complex modeling

(1)

3. 多特征融合

3.1. 三维局部二值模式(3D-LBP)

3D-LBP算法描述：对于一幅图像，是任意像素点的邻域，是中间像素点，它的灰度值被用来作为阈值。像素点邻域的其他像素点被定义为，其中是用来标记邻域像素点。中心像素点对应LBP值的计算方法如下：

(2)

(3)

LBP算法相对容易实现，因为在(2)式中LBP算法只需要中心像素与其相邻像素的灰度值差值来计算。图2左边是一幅灰度图像，右边是它的LBP图像，而在中间有两个的矩阵则是以灰度值为207的中心像素为例，详细介绍了LBP算法。中心像素作为一个阈值的函数，并在右侧的矩阵中揭示了通过中心像素与其相邻像素之间灰度值的差异来进行阈值去噪的过程。通过(2)式和(3)式，可以得到一个阈值为207的符号函数，进而得到LBP的二进制代码为00001111，其对应于的十进制数为15。

3.2. Harr小波变换

3.3. 颜色特征提取

Figure 2. In the case of a threshold of 207, showing the process of gray images into LBP images

Figure 3. Description of 3D-LBP images

Figure 4. The sketch map of Harr wavelet transform

(4)

(5)

(6)

L的取值为，计算L可以得到64柄的一维直方图，而H、S、V三个分量也通过L在一维矢量上分布开来。色调H取的权重取为8，饱和度S的权重取为2，亮度V的权重取为1，这就大大减轻了图像亮度V对检索结果的影响，而且也削弱了饱和度S对检索结果的影响，却能够对颜色分布不同的图像得到较好的检索结果。

LL的颜色特征提取还需要运用到动态金字塔理论。首先，将LL图像压缩成方阵，进行HSV空间颜色特征提取，得到64维特征。再将压缩后的方阵进行减半处理，行列分别交叉进行选取。本文将LL图像缩减3次，分别在HSV空间提取颜色特征，得到192维特征，加上原始LL图像的64维颜色特征，共256维。

3.4. 纹理特征选取

Figure 5. The sketch map of extracting the shape features through HOG descriptors after Harr wavelet transform

3.5. 多特征融合

4. 实验

Figure 6. The Precision-Recall curve

5. 结论

Image Feature Extraction Based on Complex Network and Multi-Feature Fusion Schemes Exploration in CBIR[J]. 图像与信号处理, 2015, 04(04): 101-110. http://dx.doi.org/10.12677/JISP.2015.44012

1. 1. da Costa, L.F., Rodrigues, F.A., Travieso, G., et al. (2008) Characterization of complex networks: A survey of measurements. Advances in Physics, 56, 167-242. http://dx.doi.org/10.1080/00018730601170527

2. 2. Goncalves, W.N., Silva, J.A. and Bruno, O.M. (2010) A rotation invariant face recognition method based on complex network. Proceedings of the 15th Iberoamerican Congress Conference on Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications (CIARP’10), Sao Paulo, 8-11 November 2010, 426-433. http://dx.doi.org/10.1007/978-3-642-16687-7_57

3. 3. Backes, A.R., Martinez, A.S. and Bruno, O.M. (2011) Texture analysis using graphs generated by deterministic partially self-avoiding walks. Pattern Recognition, 44, 1684-1689. http://dx.doi.org/10.1016/j.patcog.2011.01.018

4. 4. Tang, J., Jiang, B., Chang, C.C., et al. (2012) Graph structure analysis based on complex network. Digital Signal Processing, 22, 713-725. http://dx.doi.org/10.1016/j.dsp.2012.04.011

5. 5. Backes, A.R., Casanova, D. and Bruno, O.M. (2009) A complex network based approach for boundary shape analysis. Pattern Recognition, 42, 54-67. http://dx.doi.org/10.1016/j.patcog.2008.07.006

6. 6. Backes, A.R. and Bruno, O.M. (2010) Shape classification using complex network and multi-scale fractal dimension. Pattern Recognition Letters, 31, 44-51. http://dx.doi.org/10.1016/j.patrec.2009.08.007

7. 7. 郅大鹏, 等 (2013) 多模复杂网络模型的形状特征提取方法. 计算机科学与探索, 6, 570-576.

8. 8. 汤进, 等 (2013) 基于复杂网络的图像建模与特征提取方法. 计算机工程, 5, 243-252.

9. 9. Backes, A.R., Casanova, D. and Bruno, O.M. (2009) A complex network-based approach for boundary shape analysis. Pattern Recognition, 42, 54-67. http://dx.doi.org/10.1016/j.patcog.2008.07.006

10. 10. 束鑫, 吴小俊, 潘磊 (2009) 一种新的基于形状轮廓点分布的图像检索. 光电子•激光, 10, 1385-1389.

11. 11. Luo, B. and Edwin, R.H. (2001) Structural graph matching using the EM algorithm and singular value decomposition. Pattern Analysis and Machine Intelligence, 23, 1120-1136. http://dx.doi.org/10.1109/34.954602

12. 12. Tang, J., Zhang, C.Y. and Luo, B. (2006) A new approach to graph seriation. Proceedings of the International Conference on Innovative Computing, Information and Control, Beijing, 30 August 2006.

13. 13. Ojala, T., Pietikainen, M. and Harwood, D. (1994) Performance evaluation of texture measures with classification based on Kullback discrimination of distributions. Proceedings of the International Conference on Pattern Recognition, Jerusalem, 9-13 October 1994, 582-585. http://dx.doi.org/10.1109/ICPR.1994.576366

14. 14. Ojala, T., Pietikainen, M. and Harwood, D. (1996) A comparative study of texture measures with classification based on feature distributions. Pattern Recognition, 29, 51-59. http://dx.doi.org/10.1016/0031-3203(95)00067-4

15. 15. Ojala, T., Pietikainen, M. and Maenpa, T. (2002) Multiresolution gray-scale and rotation invariant texture classification with local binary patterns. IEEE Transactions on Pattern Analysis and Machine Intelligence, 24, 971-987. http://dx.doi.org/10.1109/TPAMI.2002.1017623

16. 16. Banerji, S., Verma, A. and Liu, C. (2011) Novel color LBP descriptors for scene and image texture classification. Proceedings of the 15th International Conference on Image Processing, Computer Vision, and Pattern Recognition, Las Vegas, 18-21 July 2011, 537-543.

17. 17. Liu, C. (2008) Learning the uncorrelated, independent, and discriminating color spaces for face recognition. IEEE Transactions on Information Forensics and Security, 3, 213-222. http://dx.doi.org/10.1109/TIFS.2008.923824

18. 18. Liu, C. (2006) Capitalize on dimensionality increasing techniques for improving face recognition grand challenge performance. IEEE Transactions on Pattern Analysis and Machine Intelligence, 28, 725-737. http://dx.doi.org/10.1109/TPAMI.2006.90

19. 19. Burrus, C., Gopinath, R. and Guo, H. (1998) Introduction to wavelets and wavelet transforms: A primer. Prentice-Hall, Englewood Cliffs.

20. 20. Kekre, H.B., Thepade, S.D. and Maloo, A. (2010) Query by image content using color-texture features extracted from Haar wavelet pyramid. International Journal of Computer Applications, 2, 52.

21. 21. Huu, Q.N., Thu, H.N.T. and Quoc, T.N. (2012) An efficient content based image retrieval method for retrieving images. International Journal of Innovative Computing Information and Control, 4, 2823-2836.

*通讯作者。