纺织学报 ›› 2020, Vol. 41 ›› Issue (04): 45-50.doi: 10.13475/j.fzxb.20190605606

• 纺织工程 • 上一篇    下一篇

织物纹样特征提取与匹配方法比较

汪会1, 孙洁1, 丁笑君1,2, 龙颖1, 邹奉元1,2,3()   

  1. 1.浙江理工大学 服装学院, 浙江 杭州 310018
    2.浙江理工大学 浙江省服装工程技术研究中心,浙江 杭州 310018
    3.浙江省服装个性化定制协同创新中心, 浙江 杭州 310018
  • 收稿日期:2019-06-24 修回日期:2020-01-18 出版日期:2020-04-15 发布日期:2020-04-27
  • 通讯作者: 邹奉元
  • 作者简介:汪会(1994—),女,硕士生。主要研究方向为服装数字化技术。
  • 基金资助:
    浙江省2011协同创新中心科技研发专项资助项目(17034005-F);2019年浙江省大学生科技创新活动计划项目(2019R406070);浙江理工大学2018年优秀研究生学位论文培育基金项目(2018-XWLWPY-M-04-04)

Comparison of feature extracting and matching methods for fabric patterns

WANG Hui1, SUN Jie1, DING Xiaojun1,2, LONG Ying1, ZOU Fengyuan1,2,3()   

  1. 1. School of Fashion Design & Engineering, Zhejiang Sci-Tech University, Hangzhou, Zhejiang 310018, China
    2. Zhejiang Provincial Research Center of Clothing Engineering Technology, Zhejiang Sic-Tech University,Hangzhou, Zhejiang 310018, China
    3. Zhejiang Garment Personalized Customization Collaborative Innovation Center, Hangzhou, Zhejiang 310018, China
  • Received:2019-06-24 Revised:2020-01-18 Online:2020-04-15 Published:2020-04-27
  • Contact: ZOU Fengyuan

RichHTML

10

PDF (PC)

282

摘要：

针对织物纹样自动识别过程中因尺度、旋转和褶皱等因素引起图像差异的问题,探索了复杂纹样特征的准确提取与匹配方法。以江崖海水纹样为例,采集尺度、旋转、模糊、光照、褶皱5种变化下的织物纹样图像,分别运用尺度不变特征变换(SIFT)、快速鲁棒性尺度不变特征(SURF)、二进制鲁棒不变可扩展关键点(BRISK)3种方法提取纹样局部特征,然后采用欧氏距离进行特征匹配计算,最后通过随机抽样一致算法剔除误匹配对。结果表明:采用BRISK算法的准确配对率最高,平均准确匹配率达87.10%;褶皱对织物特征匹配的影响最大,该变化下BRISK算法的鲁棒性优于SIFT和SURF算法;BRISK算法速度最快,图像平均匹配时间0.551 s;在织物纹样特征匹配中,BRISK算法比SIFT和SURF算法具有更好的适用性。

关键词: 江崖海水纹样, 特征提取, 二进制鲁棒不变可扩展关键点算法, 特征匹配, 织物纹样识别

Abstract:

In fabric pattern recognition, variations in image scaling, rotating, folding and other deformation in the sampling process cause errors, and this paper investigated the improvement of the exacting and matching method for fabric patterns. To explore the applicability of feature extracting and matching methods for complicated patterns, the river cliff water pattern was taken as experimental samples. 5 types of images of the fabric pattern feature were acquired under scaling, rotation, fuzzy, illumination and drape respectively, and fabric pattern features were extracted using scale-invariant feature transform(SIFT), speeded-up robust features (SURF), binary robust invariant scalable key-points (BRISK)algorithms. Euclidean distance method was adopted for matching calculation, eliminating false match points by random sample consensus algorithm. The results show that BRISK algorithm is the best in matching ratio, which is 87.10% on average. Folding is proven to have the greatest impacts on fabric feature matching, and the robustness of BRISK algorithm under folding change is better than SIFT and SURF algorithms. BRISK algorithmic speed is the fastest, taking an average time of 0.551 s to perform the image feature extraction and matching. In fabric pattern matching, BRISK algorithm is demonstrated better applicability than SIFT or SURF algorithms.

Key words: river cliff water pattern, feature extraction, binary robust invariant scalable key-points algorithm, feature matching, fabric pattern recognition

中图分类号: 

  • TS941

图1

江崖海水纹获取流程"

图2

3例江崖海水纹原图"

图3

图像预处理"

图4

尺度缩放时SIFT算法的错误匹配"

图5

原图与模糊图像匹配效果"

表1

褶皱状态下不同比例阈值的平均匹配结果对比"

阈值 算法 原图特
征点数		 变换后特
征点数		 所有匹
配数		 准确匹
配数		 准确匹
配率/%
0.3 SIFT 1 393 1 616 545 452 82.94
SURF 1 393 1 616 545 452 82.94
BRISK 2 639 2 113 140 128 91.43
0.4 SIFT 1 393 1 616 820 645 78.66
SURF 1 393 1 616 820 645 78.66
BRISK 2 639 2 113 305 266 87.21
0.5 SIFT 1 393 1 616 1 147 809 70.53
SURF 1 393 1 616 1 147 809 70.53
BRISK 2 639 2 113 708 585 82.63
0.6 SIFT 1 393 1 616 1 265 855 67.59
SURF 1 393 1 616 1 265 855 67.59
BRISK 2 639 2 113 1 023 889 86.90
0.7 SIFT 1 393 1 616 1 374 907 66.01
SURF 1 393 1 616 1 374 907 66.01
BRISK 2 639 2 113 1 255 1 049 83.59
0.8 SIFT 1 393 1 616 1 543 963 62.41
SURF 1 393 1 616 1 543 963 62.41
BRISK 2 639 2 113 1 571 1 177 74.92
0.9 SIFT 1 393 1 616 1 629 990 60.77
SURF 1 393 1 616 1 629 990 60.77
BRISK 2 639 2 113 2 010 1 202 59.80

表2

3种算法准确匹配率比较"

类别 k0 准确匹配率/%
SIFT SURF BRISK
尺度 0.6 35.63 52.01 80.04
旋转 0.6 82.60 81.38 87.10
模糊 0.6 48.85 51.67 85.86
光照 0.6 70.83 67.45 88.63
褶皱 0.6 51.71 52.22 66.78

表3

5种变化下不同算法的匹配时间比较"

变化 匹配时间
SIFT SURF BRISK
尺度均值 1.382 1.612 0.504
旋转均值 0.817 1.046 0.330
模糊均值 1.282 1.497 0.444
光照均值 1.593 1.106 0.621
褶皱均值 1.607 1.914 0.855
均值 1.336 1.435 0.551

图6

织物纹样"

表4

5种变化下织物纹样A、B的平均准确匹配率"

纹样名称 类别 SIFT SURF BRISK
纹样A 尺度 62.40 63.81 74.30
旋转 78.04 80.52 81.18
模糊 63.63 63.08 81.13
光照 86.93 89.18 96.17
褶皱 66.64 77.95 77.95
纹样B 尺度 50.43 47.83 84.11
旋转 84.83 86.07 93.02
模糊 43.51 44.56 88.82
光照 83.97 83.69 98.13
褶皱 66.69 66.68 88.19
[1] ZHAO H, JIA G, TAN X. Fabric pattern elements retrieval based on cosine transform[J]. Computer Aided Drafting, Design and Manufacturing, 2015,25(3):18-22.
[2] JING J, LI Q, LI P, et al. A new method of printed fabric image retrieval based on color moments and gist feature description[J]. Textile Research Journal, 2016,86(11):1137-1150.
[3] DENG D, WANG R, WU H, et al. Learning deep similarity models with focus ranking for fabric image retrieval[J]. Image and Vision Computing, 2017,70(12):11-20.
[4] 曹霞, 李岳阳, 罗海驰, 等. 基于层次匹配下多种特征融合的蕾丝花边检索方法[J]. 计算机工程与科学, 2017,39(9):1691-1699.
CAO Xia, LI Yueyang, LUO Haichi, et al. A lace retrieval method based on hierarchical matching and multiple features[J]. Computer Engineering & Science, 2017,39(9):1691-1699.
[5] LU K, ZHONG Y, LI D, et al. Cashmere/wool identification based on bag-of-words and spatial pyramid match[J]. Textile Research Journal, 2018,88(21):2435-2444.
[6] 丁笑君, 邹楚杭, 陈敬玉, 等. 畲族服装特征提取及其分布[J]. 纺织学报, 2015,36(7):110-115.
DING Xiaojun, ZOU Chuhang, CHEN Jingyu, et al. RExtraction and distribution of She nationality clothing characteristics[J]. Journal of Textile Research, 2015,36(7):110-115.
[7] 陈金广, 曹春梅, 任冰青, 等. 特征提取与匹配算法在民族服饰图案上的应用研究[J]. 丝绸, 2015,52(5):36-41.
CHEN Jinguang, CAO Chunmei, REN Bingqing, et al. Research on application of feature extraction and matching algorithms in chinese costume patterns[J]. Journal of Silk, 2015,52(5):36-41.
[8] 王业宏. 清代龙袍纹样探析[J]. 丝绸, 2009,46(7):48-51.
WANG Yehong. Analysis on the patterns of Qing dragon robes[J]. Journal of Silk, 2009,46(7):48-51.
[9] 向忠, 何旋, 钱淼, 等. 基于边缘和颜色特征的织物印花花型检索[J]. 纺织学报, 2018,39(5):137-143.
XIANG Zhong, HE Xuan, QIAN Miao, et al. Printed fabric pattern retrieval based on edge and color fea-tures[J]. Journal of Textile Research, 2018,39(5):137-143.
[10] 杨帆, 郭建华, 谭海, 等. 不同距离测度的SIFT特征描述符相似性度量比较[J]. 遥感信息, 2017,2(1):104-108.
YANG Fan, GUO Jianhua, TAN Hai, et al. Comparison of similarity measurement of SIFT feature descriptor based on different distance measure[J]. Remote Sensing Information, 2017,2(1):104-108.
[11] SCHNABEL R, WAHL R, KLEIN R. Efficient RANSAC for point-cloud shape detection[J]. Computer Graphics Forum, 2010,26(2):214-226.
[1] 孙洁, 丁笑君, 杜磊, 李秦曼, 邹奉元. 基于卷积神经网络的织物图像特征提取与检索研究进展[J]. 纺织学报, 2019, 40(12): 146-151.
[2] 吴传彬, 刘骊, 付晓东, 刘利军, 黄青松. 结合显著区域检测和手绘草图的服装图像检索[J]. 纺织学报, 2019, 40(07): 174-181.
[3] 赵树志 狄岚 何锐波. 基于改进判别性完整局部二值模式与格分割的织物瑕疵检测方法[J]. 纺织学报, 2018, 39(09): 57-64.
[4] 董蓉 李勃 徐晨. 应用积分图的织物瑕疵检测快速算法[J]. 纺织学报, 2016, 37(11): 141-147.
[5] 曹霞 李岳阳 罗海驰 蒋高明 丛洪莲. 蕾丝花边的改进型纹理特征检索方法[J]. 纺织学报, 2016, 37(06): 142-154.
[6] 黎聪 闫学娜 曾祥忠 梁猛 张莹. 应用一维傅里叶变换的剖幅区自动识别与定位[J]. 纺织学报, 2016, 37(01): 147-151.
[7] 康锋 胡洁 张华熊 周慧. 基于最大稳定极值区域的织物图案检索[J]. 纺织学报, 2015, 36(10): 156-160.
[8] 林翔宇 张华熊. 基于图像边缘特征的领带花型检索方法[J]. 纺织学报, 2015, 36(05): 123-126.
[9] 杜玉红 王加富 蒋秀明 周国庆 罗永恒. 应用聚类统计分析的棉花异纤图形检测算法[J]. 纺织学报, 2015, 36(03): 135-139.
[10] 管声启 高照元 吴宁 徐帅华. 基于视觉显著性的平纹织物疵点检测[J]. 纺织学报, 2014, 35(4): 56-0.
[11] 杨晓波. 基于GMRF模型的统计特征畸变织物疵点识别[J]. 纺织学报, 2013, 34(4): 137-142.
[12] 杨晓波. 基于自适应离散小波变换的混合特征畸变织物疵点识别[J]. 纺织学报, 2013, 34(1): 133-137.
[13] 杨晓波. 基于人工神经网络的织物疵点聚类分析[J]. 纺织学报, 2011, 32(9): 29-33.
[14] 刘素一;夏蕾. 基于元胞自动机的织物图像拼接[J]. 纺织学报, 2011, 32(1): 29-33.
[15] 蒋高平;钟跃崎;王荣武. 基于谱线特征的羊绒与羊毛的鉴别[J]. 纺织学报, 2010, 31(4): 15-19.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
京ICP备14006648号　 京公网安备11010502044800号
版权所有 © 2021 《纺织学报》编辑部
地址： 北京市朝阳区延静里中街3号主楼6层《纺织学报》杂志社　邮政编码：100025　
电话：010-65017711,65917740　传真：010-65016539　Email:fangzhixuebao@vip.126.com
本系统由北京玛格泰克科技发展有限公司设计开发