Journal of Textile Research ›› 2021, Vol. 42 ›› Issue (06): 146-152.doi: 10.13475/j.fzxb.20200905607

• Apparel Engineering • Previous Articles     Next Articles

Research on clothing collar types based on complex network extraction and support vector machine classification

XU Zengbo(), ZHANG Ling, ZHANG Yanhong, CHEN Guiqing   

  1. College of Fashion, Shanghai University of Engineering Science, Shanghai 201600, China
  • Received:2020-09-21 Revised:2021-03-03 Online:2021-06-15 Published:2021-06-28

Abstract:

In order to achieve automatic style search in clothing pattern-making, this research took the structural features of clothing collar styles as working object, using clothing round-neck images as an example. The paper described and extracted complex network features by constructing a complex network, and the support vector machine model was used to classify images of 8 types of collars. The experimental results show that the average classification accuracy of the samples as a whole is 98%, and the average classification accuracy of each category is above 96%. Among them, the average classification accuracy rate for the round collar samples is 100%. At the same time, in order to evaluate the anti-noise performance of the feature extraction algorithm, after adding a certain degree of salt and pepper noise and Gaussian noise to the image of the original sample library, the overall classification accuracy of the sample fluctuates around 80%, indicating that the support vector machine classification method is suitable for image recognition with a certain degree of noise. To conclude, the extraction and classification accuracy of clothing collar research based on complex network extraction and support vector machine classification is high, and the classification results are relatively stable.

Key words: complex network, feature extraction, collar type classification, support vector machine model, clothing design

CLC Number: 

  • TS941.2

Tab.1

Classification of image feature extraction methods"

分类 描述
幅度特征 图像像素灰度值、三色值、频谱值等表示的幅值特征是最基本的图像特征
统计特征 将图像当作一个二维随机过程的一次实现,如图像的灰度直方图、均值、方差、偏差、峰度、能量、协方差等
变化系数
特征
对于具有唯一性的任何变换,其变换域系数所决定的亮度图像,都和原空间域图像是等价的
边界特征 以灰度和三色值表示的亮度突变或者断续,成为图像中的亮度边界点
拓扑结构
特征
它是只要图形不撕裂或折叠,是不受任何畸变影响的结构特点,这些特点既不受距离的影响,也不受基于距离测量的其他特点的影响
几何形状
特征
目标物体的几何特征,如周长、面积、弦长等

Fig.1

Initial network construction. (a) Target part;(b) Target edge; (c) Uniform sampling; (d) Initial network"

Fig.2

Complex network model evolution with different thresholds"

Fig.3

Trend chart of different characteristics with threshold.(a) Maximum and average thresholds;(b)Average joint degree, entropy and energy threshold;(c)Local and global clustering coefficients and thresholds of average path length"

Tab.2

Type and number of experimental samples"

衣领款式 类别标签 样本总数 训练样本量 测试样本量
圆领 1# 60 45 15
V领 2# 60 45 15
方形领 3# 60 45 15
立领 4# 60 45 15
衬领 5# 60 45 15
平驳领 6# 60 45 15
连驳领 7# 60 45 15
戗驳领 8# 60 45 15

Tab.3

Type and number of experimental samples"

核函数类型 平均分类准确率/% svmtrain参数
linear 97.50 ‘-c2-g1-t0’
polynomial 96.92 ‘-c2-g1-t1’
radial basis function 95.58 ‘-c2-g1-t2’
sigmoid 8.92 ‘-c2-g1-t3’

Fig.4

Different category classification map based on SVM.(a) 2# and 3#;(b) 1# and 4#;(c) 3# and 8#;(d) 1#, 5# and 6# "

Tab.4

Type and number of experimental samples"

实验次数 圆领 V领 方形领 立领 衬衫领 平驳领 连驳领 戗驳领 整体准确率
1 1.00 1.00 1.00 1.00 1.00 0.93 0.87 0.95 0.97
2 1.00 0.93 1.00 1.00 1.00 1.00 1.00 1.00 0.99
3 1.00 1.00 1.00 1.00 1.00 0.94 1.00 0.94 0.98
4 1.00 0.95 1.00 1.00 1.00 0.88 1.00 0.92 0.97
5 1.00 1.00 1.00 0.94 0.92 0.92 1.00 1.00 0.98
6 1.00 1.00 1.00 1.00 0.93 0.94 0.86 1.00 0.97
7 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
8 1.00 0.90 1.00 0.94 1.00 1.00 1.00 0.88 0.97
9 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.92 0.99
10 1.00 1.00 0.93 1.00 1.00 1.00 0.94 1.00 0.98
平均准确率 1.00 0.98 0.99 0.99 0.98 0.96 0.97 0.96 0.98

Fig.5

Trend chart of different characteristics with threshold.(a) Original image;(b) Noise 10%; (c) Noise 15%;(d) Noise 20%"

Fig.6

Add Gaussian noise images of different densities. (a) Original image; (b) Variance 0.01; (c) Variance 0.05; (d) Variance 0.1"

Fig.7

Add classification accuracy of different densities of salt and pepper noise images. (a) Overall classification accuracy of salt and pepper;(b) Average accuracy of salt and pepper classification"

Fig.8

Add different variance of Gaussian noise image classification accuracy.(a) Gaussian overall classification accuracy;(b) Gaussian average classification accuracy"

Fig.9

Comparison of classification accuracy of this method with Hu invariant moments and HOG.(a) Overall classification accuracy of the three methods;(b) Average classification accuracy of three methods"

[1] ANDRÉ Ricardo Backes, CASANOVA D, BRUNO O M. A complex network-based approach for boundary shape analysis[J]. Pattern Recognition, 2009, 42(1):54-67.
doi: 10.1016/j.patcog.2008.07.006
[2] 阮瑞, 江波, 汤进, 等. 改进复杂网络模型的形状特征提取[J]. 中国图象图形学报, 2014, 19(9):1332-1337.
RUAN Rui, JIANG Bo, TANG Jin, et al. Improved shape feature extraction of complex network models[J]. Journal of Image and Graphics, 2014, 19(9):1332-1337.
[3] 汤进, 郅大鹏, 江波, 等. 基于有向复杂网络模型的形状描述与识别[J]. 计算机辅助设计与图形学学报, 2014, 26(11):2039-2045.
TANG Jin, ZHI Dapeng, JIANG Bo, et al. Shape description and recognition based on directed complex network model[J]. Journal of Computer Aided Design and Graphics, 2014, 26(11):2039-2045.
[4] 李咏豪. 基于KNN有向复杂网络的图像轮廓识别[J]. 计算机时代, 2019(6):31-33,36.
LI Yonghao. Image contour recognition based on KNN directed complex network[J]. Computer Times, 2019(6):31-33,36.
[5] 李东. 基于数字图像处理的服装款式识别方法研究[D]. 上海: 东华大学, 2017:27-32.
LI Dong. Research on clothing style recognition method based on digital image processing[D]. Shanghai: Donghua University, 2017:37-32.
[6] AN L X, LI W. An integrated approach to fashion flat sketches classification[J]. International Journal of Clothing Science & Technology, 2014, 26(21):346-366.
[7] 张铮, 徐超, 任淑霞, 等. 数字图像处理与机器视觉 Visual C++与Matlab实现[M]. 2版. 北京: 人民邮电出版社, 2014:23-26.
ZHANG Zheng, XU Chao, REN Shuxia, et al. Digital image processing and machine vision visual C++ and matLab implementation[M]. 2rd ed. Beijing: People's Posts and Telecommunications Press, 2014:23-26.
[8] 赵苗苗. 基于内容的服装图像检索系统的研究与应用[D]. 电子科技大学, 2015:28-35.
ZHAO Miaomiao. Research and application of content-based clothing image retrieval system[D]. University of Electronic Science and Technology of China, 2015:28-35.
[9] 宋明秋, 曹晓芸. 基于敏感特征的网络钓鱼网站检测方法[J]. 大连理工大学学报, 2013, 53(6):903-907.
SONG Mingqiu, CAO Xiaoyun. Phishing website detection method based on sensitive features[J]. Journal of Dalian University of Technology, 2013, 53(6):903-907.
[10] 谢建春. 基于改进 Hausdorff 距离的图像匹配快速算法[J]. 电光与控制, 2012, 19(8):34-27.
XIE Jianchun. Fast image matching algorithm based on improved Hausdorff distance[J]. Electro-Optics and Control, 2012, 19(8):34-27.
[11] 黄艳华, 孙文磊. 二维视图特征自动识别的新途径[C]// 先进制造技术论坛暨第二届制造业自动化与信息化技术交流会论文集. 北京: 中国机械工程学会, 2003:3.
HUANG Yanhua, SUN Wenlei. A new approach for automatic recognition of two-dimensional view fea-tures[C]// Advanced Manufacturing Technology Forum and the Second Manufacturing Automation and Information Technology Exchange Conference Proceedings. Beijing: Chinese Mechanical Engineering Society, 2003:3.
[12] KHALIL M I, BAYOUMI M M. Affine invariants for object recognition using the wavelet transform[M]. Amsterdam: Elsevier Science Inc, 2002:56-79.
[13] 耿庆田, 赵浩宇, 于繁华, 等. 基于改进HOG特征提取的车型识别算法[J]. 中国光学, 2018, 11(2):174-181.
GENG Qingtian, ZHAO Haoyu, YU Fanhua, et al. Vehicle recognition algorithm based on improved HOG feature extraction[J]. China Optics, 2018, 11(2):174-181.
doi: 10.3788/co.
[1] ZHENG Xiaoping, LIU Chengxia. Objective evaluation of woven garment bagging performance by laser measurement [J]. Journal of Textile Research, 2021, 42(05): 143-149.
[2] ZHANG Heng. Design method of lapel collar structure based on structure model of lapel collar looseness [J]. Journal of Textile Research, 2020, 41(11): 128-135.
[3] YANG Juan, ZHANG Yuanpeng. Garment ornamenting craft classification model for knowledge graph on clothing design [J]. Journal of Textile Research, 2020, 41(08): 95-100.
[4] WANG Hui, SUN Jie, DING Xiaojun, LONG Ying, ZOU Fengyuan. Comparison of feature extracting and matching methods for fabric patterns [J]. Journal of Textile Research, 2020, 41(04): 45-50.
[5] . Fabric defect inspection based on modified discriminant complete local binary pattern and lattice segmentation [J]. Journal of Textile Research, 2018, 39(09): 57-64.
[6] . Analysis on creative design of Lujin [J]. JOURNAL OF TEXTILE RESEARCH, 2017, 38(06): 58-63.
[7] . Fast fabric defect detection algorithm based on integral image [J]. JOURNAL OF TEXTILE RESEARCH, 2016, 37(11): 141-147.
[8] . Lace retrieval method based on improved texture feature [J]. JOURNAL OF TEXTILE RESEARCH, 2016, 37(06): 142-154.
[9] . Automatic recognition and positioning method of fabric slitting zone using one-dimensional Fourier transform [J]. JOURNAL OF TEXTILE RESEARCH, 2016, 37(01): 147-151.
[10] . Research progress and development trend of wearable medical monitoring clothing [J]. Journal of Textile Research, 2015, 36(06): 162-168.
[11] . Algorithm for pattern detection of cotton foreign fibers based on cluster statistic analysis [J]. Journal of Textile Research, 2015, 36(03): 135-139.
[12] . Defect detection of plain weave based on visual saliency mechanism [J]. JOURNAL OF TEXTILE RESEARCH, 2014, 35(4): 56-0.
[13] . Fabric defect detection of statistic aberration feature based on GMRF model [J]. JOURNAL OF TEXTILE RESEARCH, 2013, 34(4): 137-142.
[14] . Research of mixture feature aberrance fabric defect recognition based on self-adaptive disperse wavelet transform [J]. JOURNAL OF TEXTILE RESEARCH, 2013, 34(1): 133-137.
[15] . Fabric Defect Clustering Analysis based on Artificial Neural Network [J]. JOURNAL OF TEXTILE RESEARCH, 2011, 32(9): 29-33.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!