基于BoF模型的多特征融合纹理图像分类

汪宇玲; 黎明; 李军华; 张聪炫; 陈昊

doi:10.13700/j.bh.1001-5965.2017.0720

基于BoF模型的多特征融合纹理图像分类

doi: 10.13700/j.bh.1001-5965.2017.0720

汪宇玲^{1, 2},
黎明^{1, 3, ,},
李军华³,
张聪炫³,
陈昊³

1.
南京航空航天大学自动化学院, 南京 211106
2.
东华理工大学信息工程学院, 南昌 330013
3.
南昌航空大学图像处理与模式识别江西省重点实验室, 南昌 330063

基金项目:

国家自然科学基金 61262019

国家自然科学基金 61402102

航空科学基金 2015ZC56009

江西省优势创新团队项目 20113BCB24009

江西省自然科学基金 20151BAB207042

江西省教育厅科技项目 GJJ160554

江西省教育厅科技项目 GJJ170432

详细信息

作者简介:
汪宇玲女, 博士研究生, 副教授。主要研究方向:图像处理、模式识别

黎明男, 博士, 教授, 博士生导师。主要研究方向:图像处理、智能计算

通讯作者:
黎明, E-mail:liming@nchu.edu.cn

中图分类号: TP391.4
计量
- 文章访问数: 1034
- HTML全文浏览量: 247
- PDF下载量: 412
- 被引次数: 0
出版历程
- 收稿日期: 2017-11-21
- 录用日期: 2017-12-29
- 网络出版日期: 2018-09-20

Texture image classification based on BoF model with multi-feature fusion

WANG Yuling^{1, 2},
LI Ming^{1, 3
, ,},
LI Junhua³,
ZHANG Congxuan³,
CHEN Hao³

1.
College of Automation Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China
2.
School of Information Engineering, East China University of Technology, Nanchang 330013
3.
Key Laboratory of Jiangxi Province for Image Processing and Pattern Recognition, Nanchang Hangkong University, Nanchang 330063, China

Funds:

National Natural Science Foundation of China 61262019

National Natural Science Foundation of China 61402102

Aeronautical Science Foundation of China 2015ZC56009

Jiangxi Advantage Innovative Research Team 20113BCB24009

Jiangxi Natural Science Foundation, China 20151BAB207042

Science and Technology Project of Education Department of Jiangxi Province GJJ160554

Science and Technology Project of Education Department of Jiangxi Province GJJ170432

More Information

Corresponding author: LI Ming, E-mail:liming@nchu.edu.cn

摘要

摘要:
针对特征词袋（BoF）模型缺乏空间和几何信息，对纹理图像内容表达不明显等问题，提出一种基于BoF模型的多特征融合纹理分类算法。将灰度梯度共生矩阵（GGCM）和尺度不变特征转换（SIFT）融合特征作为纹理图像的区域特征描述，通过动态权重鉴别能量分析进行最优参数特征选择，并用BoF量化纹理特征，使用支持向量机对图像进行训练和预测，得出分类结果。实验结果表明，本文算法对有旋转扭曲的纹理、边缘模糊纹理、有光照变化的纹理及杂乱纹理等均能取得较好的分类效果，相对于传统BoF模型及凹凸划分（CCP）方法等算法在UIUC纹理库上的分类正确率均有不同程度的提高，平均分类正确率分别提高12.8%和7.9%，说明本文算法针对纹理图像分类具有较高的精度和较好的鲁棒性。
- 纹理分类 /
- 多特征融合 /
- 特征词袋(BoF) /
- 灰度梯度共生矩阵(GGCM) /
- 尺度不变特征转换(SIFT)
Abstract:
The obvious shortcomings of bag of feature (BoF) model are lack of spatial and geometric information in image representation, and poor description of the content of texture image. To solve these problems, we proposed a texture image classification method based on BoF model with multi-feature fusion. The method fuses gray gradient co-occurrence matrix (GGCM) and scale-invariant feature transform (SIFT) as the basic feature description of texture image, uses a dynamic weight to identify energy analysis for the optimal parameter feature selection, quantifies texture feature by BoF, then applies support vector machine to train and predict the image, and finally obtains the classification results. The experimental results show that the proposed method has better performance of texture classification into rotated texture, twisted texture, edge fuzzy texture, light changing texture, messy texture, etc. The average classification accuracy of the proposed method on the UIUC texture database increases by 12.8% and 7.9% respectively compared with the conventional BoF model and concave-convex partition (CCP) methods, which indicates that the proposed method has higher accuracy and better robustness for texture image classification.
- texture classification /
- multi-feature fusion /
- bag of feature (BoF) /
- gray gradient co-occurrence matrix (GGCM) /
- scale-invariant feature transform (SIFT)

HTML全文

图 1 生成特征词典

Figure 1. Feature dictionary generation

下载: 全尺寸图片幻灯片

图 2 生成图像BoF向量

Figure 2. Image BoF vector generation

下载: 全尺寸图片幻灯片

图 3 基于BoF和多特征融合的纹理图像分类算法流程图

Figure 3. Flowchart of texture image classification algorithm based on BoF and multi-feature fusion

下载: 全尺寸图片幻灯片

图 4 实验用部分纹理图像示例

Figure 4. Partial texture image samples in experiments

下载: 全尺寸图片幻灯片

图 5 单一特征BoF模型与PBoFF算法分类正确率比较

Figure 5. Comparison of classification accuracy between single feature BoF model and PBoFF algorithm

下载: 全尺寸图片幻灯片

图 6 本文算法与文献[23]的结果比较

Figure 6. Comparison of results between proposed algorithm and Ref.[23]

下载: 全尺寸图片幻灯片

表 1 单一特征及其结合BoF模型分类正确率比较

Table 1. Comparison of classification accuracy of single feature and its BoF model

%
特征	分类正确率
特征	GLCM	GGCM
单一特征	75.8	87.8
结合BoF模型	78.3	90.6

下载: 导出CSV

表 2 特征词典容量对PBoFF算法分类性能的影响

Table 2. Effect of feature dictionary capacity on classification performance of PBoFF algorithm

词典容量	分类正确率/%	时间/s
58	94.4	3.688 821
116	96.5	5.364 589
290	98.4	15.394 532
580	98.1	33.014 725

下载: 导出CSV

参考文献(23)

[1]	周莉, 胡德文, 周宗潭.综合结构和纹理特征的场景识别[J].中国科学:信息科学, 2012, 42(6):687-702. ZHOU L, HU D W, ZHOU Z T.Scene recognition combining structural and textural features[J].Scientia Sinica Informationis, 2012, 42(6):687-702(in Chinese).
[2]	刘丽, 匡纲要.图像纹理特征提取方法综述[J].中国图象图形学报, 2009, 14(4):622-635. LIU L, KUANG G Y.Overview of image textural feature extraction methods[J].Journal of Image and Graphics, 2009, 14(4):622-635(in Chinese).
[3]	杨俊俐, 姜志国, 周全, 等.基于条件随机场的遥感图像语义标注[J].航空学报, 2015, 36(9):3069-3081. YANG J L, JIANG Z G, ZHOU Q, et al.Remote sensing image semantic labeling based on conditional random field[J].Acta Aeronautica et Astronautica Sinica, 2015, 36(9):3069-3081(in Chinese).
[4]	MULLER S J, NIEKERK A V.Identification of WorldView-2 spectral and spatial factors in detecting salt accumulation in cultivated fields[J].Geoderma, 2016, 273:1-11. doi: 10.1016/j.geoderma.2016.02.028
[5]	曹琼, 郑红, 李行善.一种基于纹理特征的卫星遥感图像云探测方法[J].航空学报, 2007, 28(3):661-666. doi: 10.3321/j.issn:1000-6893.2007.03.029 CAO Q, ZHENG H, LI X S.A method for detecting cloud in satellite remote sensing image based on texture[J].Acta Aeronautica et Astronautica Sinica, 2007, 28(3):661-666(in Chinese). doi: 10.3321/j.issn:1000-6893.2007.03.029
[6]	ZHANG X, YANG Y H, HAN Z, et al.Object class detection:A survey[J].ACM Computing Survey, 2013, 46(1):1-53.
[7]	梁晔, 于剑, 刘宏哲.基于BoF模型的图像表示方法研究[J].计算机科学, 2014, 41(2):36-44. doi: 10.3969/j.issn.1002-137X.2014.02.008 LIANG Y, YU J, LIU H Z.Study of BoF model based image representation[J].Computer Science, 2014, 41(2):36-44(in Chinese). doi: 10.3969/j.issn.1002-137X.2014.02.008
[8]	SANTOS F L C D, PACI M, NANNI L, et al.Computer vision for virus image classification[J].Biosystems Engineering, 2015, 138(1):11-22.
[9]	NANNI L, MELUCCI M.Combination of projectors, standard texture descriptors and bag of features for classifying images[J].Neurocomputing, 2016, 173(1):1602-1614.
[10]	YUAN Y, LI B, MENG Q H.Improved bag of feature for automatic polyp detection in wireless capsule endoscopy images[J].IEEE Transactions on Automation Science & Engineering, 2015, 13(2):1-7.
[11]	CAO J, YU S, LIU H, et al.Hand posture recognition based on heterogeneous features fusion of multiple kernels learning[J].Multimedia Tools & Applications, 2015, 75(19):1-20.
[12]	JOACHIMS T.A probabilistic analysis of the rocchio algorithm with TFIDF for text categorization[C]//International Conference on Machine Learning, 1997: 143-151.
[13]	MAJI S, BERG A C, MALIK J.Classification using intersection kernel support vector machines is efficient[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition DBLP.Piscataway, NJ: IEEE Press, 2008: 1-8.
[14]	JURIE F, TRIGGS B.Creating efficient codebooks for visual recognition[C]//Proceedings of 10th IEEE International Conference on Computer Vision.Piscataway, NJ: IEEE Press, 2005: 604-610.
[15]	罗会兰, 郭敏杰, 孔繁胜.一种基于多级空间视觉词典集体的图像分类方法[J].电子学报, 2015, 43(4):684-693. doi: 10.3969/j.issn.0372-2112.2015.04.009 LUO H L, GUO M J, KONG F S.An image classification method based on multiple level spatial visual dictionary ensemble[J].Acta Electronica Sinica, 2015, 43(4):684-693(in Chinese). doi: 10.3969/j.issn.0372-2112.2015.04.009
[16]	JULESZ B.Textons, the elements of texture perception, and their interactions[J].Nature, 1981, 290(5802):91-97. doi: 10.1038/290091a0
[17]	洪继光.灰度-梯度共生矩阵纹理分析方法[J].自动化学报, 1984, 10(1):22-25. HONG J G.Gray level-grandient cooccurrence matrix[J].Acta Automatica Sinica, 1984, 10(1):22-25(in Chinese).
[18]	黎明, 鲁方波, 陈昊.基于QWT和GLCM的多特征双重加权纹理分割[J].模式识别与人工智能, 2014, 27(3):263-271. LI M, LU F B, CHEN H.Dual weighted multi-feature texture segmentation based on QWT and GLCM[J].Pattern Recognition and Artificial Intelligence, 2014, 27(3):263-271(in Chinese).
[19]	LOWE D.Distinctive image features from scale-invariant keypoints[J].International Journal of Computer Vision, 2004, 60(2):91-110. doi: 10.1023/B:VISI.0000029664.99615.94
[20]	CHANG C C, LIN C J.LIBSVM:A library for support vector machines[J].ACM Transactions on Intelligent Systems and Technology, 2011, 2(3):1-25.
[21]	LENG L, ZHANG J, XU J, et al.Dynamic weighted discrimination power analysis in DCT domain for face and palmprint recognition[C]//International Conference on Information and Communication Technology Convergence.Piscataway, NJ: IEEE Press, 2010: 467-471.
[22]	HOSSAIN S, SERIKAWA S.Texture databases-A comprehensive survey[J].Pattern Recognition Letters, 2013, 34(15):2007-2022. doi: 10.1016/j.patrec.2013.02.009
[23]	毋小省, 朱世松, 孙君顶, 等.基于凹凸局部二值模式的纹理图像分类[J].光电子·激光, 2014, 25(8):1627-1634. WU X S, ZHU S S, SUN J D, et al.Concave-convex local binary feature for rotation invriant texture classification[J].Journal of Optoelectronics·Laser, 2014, 25(8):1627-1634(in Chinese).