APP下载

畲族传统服装设色关联规则分析

2023-07-04曹竟文贾静徐平华林瑞冰孙晓婉

丝绸 2023年4期
关键词:关联规则可视化

曹竟文 贾静 徐平华 林瑞冰 孙晓婉

摘要: 为清晰阐释畲族传统服装设色分布及其关联规则,文章利用图像分析技术解析意象色彩配色关系。以散居于浙、赣、闽三地畲族为例,对田野调查采集的150幅典型传统服装图像进行色彩解析。通过利用自适应聚类机制提取意象色彩,分别构建三地服装色彩谱系;设计基于向量集的Apriori算法,解析畲族服装意象色之间的多元配色关系。实验结果表明,浙、赣、闽三地畲族服装用色集中于黑、红、蓝等八色,主色较为接近;最小支持度优选0.2时,能够有效区分三地服装多元配色差异。其中浙江地区二元配对色组、江西地区三元配对色组表色相对丰富。改进后的算法配色规则输出平均耗时0.032 s,能够快速解析畲族服装设色关联规则,为类似传统服装色彩分析和再生设计提供方法参考。

关键词: 设色关系;关联规则;自适应聚类;色彩解析;色彩谱系;可视化

中图分类号: TS941.2;TP391.7

文献标志码: A

文章编号: 1001-7003(2023)04-0100-07

引用页码:

041201

DOI: 10.3969/j.issn.1001-7003.2023.04.013(篇序)

纺织品服装同质化、供需错配、设计决策迟缓等问题的长期显现,成为产业高质量发展的瓶颈制约之一。随着“东方美学”“新时尚”等消费势态的激活,再生传统、民族、本土化的服饰精神和文化内涵,为消费增长开辟了新思路和新途径。

畲族是分布在中国东南地区的散杂居少数民族,其服饰色彩独具特色。相关学者从审美特征[1-3]、文化变迁[4],宗教信仰[5]及染色工艺发展[6]等方面对畲族传统服饰进行评述。夏帆等[2]以畲族相关史料典藏为线索,提出畲族服装典型样式中青、蓝、黑为主要色系;陈敬玉[4]认为畲族在历史迁徙过程中与周边民族的融合,使其在各地形成特有的服饰风貌外观;吴微微等[5]认为畲族盛装中所呈现的对红色与黑色的尊崇,反映出畲族先民对太阳、火与黑暗的自然崇拜;段婷[6]则从面料印染角度出发,认为清代及后期畲族服饰色彩受到蓝染工艺的影响,逐渐形成“衣尚青蓝色”的服饰特色。当前对畲族服装设色多侧重于感性认知,缺乏系统性量化分析和地域性比对。近年来,图像分析技术逐步应用至色彩解析领域。徐平华等[7]、Hagtvedt等[8]量化分析各民族服饰所提取出的主色;刘肖健等[9-10]依托改进的色彩网络模型简洁表达色彩量化元素;徐明慧等[11]针对品牌服装构建二元配色关系模型,但在意象色多元组合方面未作深入探讨。

为此,本文重点以畲族传统服装为例,利用图像分析技术,构建浙、赣、闽三地畲族服装色彩谱系;使用改进的Apriori算法,挖掘意象色多元配色关联关系,为传统服装色彩再生设计提供配色基准。

1 设色关联规则挖掘

关联分析又称关联挖掘,是对信息载体内对象集合间频繁模式的解析。对于服装设色而言,针对批量服装图像色彩间配色关联性,利用关联规则算法挖掘其内在赋色机制。

1.1 服装用色数据集

为构建系列服装图像用色数据集,对田野考察[12]获得的畲族服装图像,按照浙、赣、闽三地进行归类。每个地区筛选了50幅代表性服装样本,三地共计150幅,所涉上装为右衽大襟衣、下装为筒裙或长裤等款式。对于含背景、噪声的图像,首先采用GrabCut[13]、高斯滤波[14]、伽马光照自适应校正[15]等算法对其进行预处理,仅保留服装主体内容,背景则采用纯白色标记。

服装用色基础数据取自序列服装图像,在批量处理服装样本图像时,各样本设色存在差异。在基础数据集构建阶段,若采用常规K-means聚类,强制统一各样本色彩聚类簇数,容易导致提色偏差。因此,本文采用二分K-均值自适应聚类[16],自适应提取每幅服装色彩。在HSV色彩空间下,对序列样本主色进行提取;在此基础上,横向比对地域差异时,采用K-means算法进行二次聚类,获得各地区服装意象色。以浙江地区畲族服装样本为例,最终获得如图1所示的首次聚类提取色和二次聚类意象色。

1.2 关联规则构造

设色规则挖掘,是对色彩融合图中超过一定阈值的配对色组,如二元、三元、四元等共现色组,进行关联度解析。当前关联规则挖掘算法中,Apriori算法[17]常用于挖掘数据关联规则,以找出数据值频繁出现的组合及其关联关系。针对该算法中的连接和修剪耗时长等缺陷,相关学者提出如FP-Growth[18]、DHP[19]和频繁闭项集法[20]等改进算法,但当数据维数较大时,运行效率较低。为快速挖掘频繁项集,本文提出了一种基于布尔矩阵运算的Apriori改进算法。

算法主要包括两个模块,一是寻找频繁项集的函数模块,评价指标为支持度,计算如下式所示:

Support(A,B)=P(A∪B)=NA,BN(1)

式中:P(A∪B)表示A、B项同时出现的比率,NA,B为A、B項同时出现的次数,N为样本数。

置信度反映了当A出现时,B出现的概率大小,如果置信度为100%,则表明A出现时必然伴随着B出现的情况。

另一模块是探索关联规则的函数模块,指标为置信度,计算如下式所示:

Confidence(AB)=P(A|B)=NA,BNA(2)

式中:P(A|B)为条件概率,表示当A出现时,B出现的概率;NA为A出现的次数。

由于置信度A→B与B→A在色彩关联规则挖掘中意义相同,为有效减少程序计算量,算法仅考虑支持度的影响。

1.3 关联规则挖掘

将上述提取色聚类所对应的意象色标记结果构建为布尔型矩阵Dn×k,下标n为图像样本数;k为二次聚类K-means设定的聚类中心数,即指定的意象色彩数。矩阵元素Dij表达如下式所示:

Dij=1I→i=Cj0I→i≠Cj(3)

式中:1≤i≤n,1≤j≤k;Cj为第j个聚类中心。

当第i幅图像中存在提取色二次聚类归为第j意象色时,Dij为1;否则,置为0。如图2所示,当k取值为8时,第1幅样本图像存在与意象色色号1、4、5、7、8相似的颜色,则对应至矩阵首行相应元素值为1,其余为0。类似地,计算矩阵Dn×k中其他元素值。

频繁项集采用与操作运算,如下式所示:

Fjt=Dj∧Dt=d1j∧d1td2j∧d2tdnj∧dnt, j,t∈(1,k)(4)

式中:Dj、Dt分别为矩阵任意两列数据项,由此计算j、t二元配对色的支持度Fjt。

计算如下式所示:

Support(Fjt)=1n∑ni=1(dij∧dit)(5)

类似地,增加公式与操作项,完成多元色组支持度的计算。

具体步骤为:通过式(3)构建色彩聚类结果对应的布尔矩阵Dn×k,根据式(5)相应地生成二元配对色组频繁项集、三元配对色组频繁项集,至多元配对色组频繁项集。当不再产生满足最小支持度的频繁项集,终止计算。

2 畲族服装设色实证分析

2.1 畲族传统服装色彩构成分析

文献[2,21-22]对畲族服装用色进行解读,指出常见色主要为黑、蓝、青、红、黄、赭、绿、灰8色。为具象化表述浙、赣、闽三地畲族服装色彩构成情况,本文采用图形化方式展示意象色分布、占比及其十六进制色值。实验中二次聚类数k同样设为8,结果如图3所示。

由H-S色环中颜色落点可以看出,三地意象色主要表现为黑、红、蓝等色,与文献[2,21-22]基本一致,但分布存在着一定的差异。浙江、江西地区畲族服装意象色多数落点在红、

紫、蓝象限,福建地区则主要落点在红、黄、青象限。此外,意象色占比排序同样存在一定的差异,若以占比50%内意象色为主色,浙江地区主色为黑、蓝和黄;江西为黑、灰和红;福建则为黑、青和黄色。

该方法直观地展示了不同地区畲族服装色彩分布及其差异。进一步地论证了畲族虽经迁徙,但用色仍保持了相对稳定,并随着与本土民俗的融合,设色形态上适度演化,形成当前的地域特征。

2.2 关联规则支持度阈值选择

关联规则中支持度阈值的设定,直接影响到配对色组解析数量。支持度阈值范围在0~1,阈值越大,解析的种类越少;反之,输出的解析种类增多。为了横向比较不同地区畲族服装设色规则,选择有效的支持度阈值,本文对不同阈值下关联规则数进行比较分析。实验中,以0.1为间隔,解析了0.1~1.0不同阈值下配对色组关系,结果如图4所示。

总体来看,当阈值由0.1逐步增大至1.0时,二元、三元、四元、五元配对数逐渐递减。当阈值增大至0.5时,浙江地区配对数均为0,江西和福建地区仅存二元配对色组;类似地,阈值为0.3、0.4时,配对类较少,不利于区分三地配色关系。而当阈值为0.1时,配对关系数过多,不利于设计人员观测内在核心关系。当阈值设置为0.2时,二元、三元规则数量适中,能够有效区分不同地区的畲族服装设色关系。因此,实验中支持度阈值设置为0.2,用以进一步地分析不同地区畲族服装设色关系。

2.3 不同地区畲族服装设色关系解析

为厘清畲族服装色彩搭配关系和运用机制,区分不同地区设色关系差异,本文对其关联关系作进一步探析。

表1显示了三地畲族服装二元配色关系及其支持度。其中,浙江地区双色规则有18组、江西地区16组、福建地区14组,各配对色组支持度数值按序排列。

为可视化呈现三地畲族服装二元配色规则,输出形式设置为:线段两端连接配色对,线段粗细表示支持度大小,连线越粗即支持度越高,即二色共现频次越高,结果如图5所示。

由表1及图5可知,三地畲族服装色彩配对关系中,浙江地区居多,用色灵活多样,配色形式丰富。福建地区较为简洁,且其F8号色未出现于配色关系中,说明其与常用色搭配使用概率低于20%。

若以支持度不低于0.4配對色组为高频配对,则浙江地区畲族服装用色Z1-Z2、Z1-Z3、Z1-Z5和Z1-Z6为高频配对,高占比Z1色号多与蓝、黄和红色相搭配呈现;江西地区高频配对色组仅为J1-J2,支持度为0.52,即江西地区畲族服装常以黑、灰两色高占比搭配出现频次高于50%,用色深沉质朴;福建地区高频配对色为F1-F3、F1-F5与F1-F4三组。

此外,浙江地区畲族服装还存在Z4、Z5、Z6与Z8四种色号交织低频配对,玫红、土红、深红等深浅不一的搭配,服装图案中以红色、水红色为主、粉色为辅的色彩搭配,黑色则协调统一所有色彩;通过色彩搭配形成区域性的视觉中心点,蕴含着一定的服装美学原则。福建地区中F3-F6、F4-F5与F4-F7配对色组支持度均为0.2,即福建地区畲族服装存在少量红、黄等点缀色作亮丽装饰配色。

表2为三地畲族服装三元配色关系及其支持度,其中浙江地区三色规则有3组、江西地区7组、福建地区5组。图6为三地畲族服装三元配色关系,三角形三条边连接三种配对色,灰色越深表示支持度越高,即畲族服装搭配中三色共现频次越高。

由表2及图6可知,江西地区畲族服装三元色配对关系较为丰富,福建次之,浙江最少。其中,浙江地区中Z1-Z2-Z3配对色搭配频次较高,为高占比黑—蓝—黄对比色搭配。江西地区存在J1、J2号色与中低占比色搭配情况,即江西地区畲族服装三元配色多存在黑、灰作主色,其余常用色点缀情况。福建地区中F1-F3-F5配对色为三地中三元配对频次最高,支持度高达0.35,其畲族服装整体呈现黑、青、不同深浅黄色三色相互配对的对比色搭配。

实验中,算法测试用计算机配置为:处理器AMD 3.59 GHz,机带RAM为8.0 GB,利用Python编写的关联规则挖掘算法,配色规则平均耗时0.032 s。

3 结 论

本文利用图像分析技术,对田野调查采集的畲族服装图像进行色彩解析。采用二分自适应聚类提取图像色彩数据,再通过两次聚类构建浙、赣、闽三地畲族服装意象色。利用改进的Apriori算法解析服裝设色规则,以可视化方式阐释不同地区设色形态和关联关系。

实验结果表明,三地畲族服装量化色与当前文献记载的畲族服装用色基本吻合,色相整体呈现为黑、红、蓝等色。二元配色关系中,各地区畲族服装色彩配对色组存在一定差异。浙江地区畲族服装存在深浅不一的同类色交织搭配情形,呈现出丰富层次感;江西地区用色深沉朴素,其中黑灰搭配频率大于50%;福建地区则存在少量红、黄等点缀色搭配情况。三元配色关系中,江西地区畲族服装配对较为丰富,浙江地区较为简洁,福建地区整体呈现黑、青、深黄与浅黄相互配对关系。上述解析的三个地区畲族服装设色关系,具象化表现出服装用色规律和配色逻辑,有助于实现对畲族服装色彩的数字化保护。此外,该方法客观、可视化的方式表征服装用色机制,为今后畲族服装色彩活化设计应用提供了赋色依据,也进一步为系统研究同类传统服装色彩提供方法参考。

参考文献:

[1]张馨翌, 冯玲玲, 李强. 畲族女装服斗造型及纹样特征研究[J]. 丝绸, 2020, 57(1): 67-75.

ZHANG Xinyi, FENG Lingling, LI Qiang. Study on the patterns and shapes of front placket decoration of She minority womens wear[J]. Journal of Silk, 2020, 57(1): 67-75.

[2]夏帆, 蔡建梅, 雷红香. 畲族服装典型样式及特征探究[J]. 丝绸, 2017, 54(4): 43-51.

XIA Fan, CAI Jianmei, LEI Hongxiang. Exploration on typical style and characteristics of She minority clothing[J]. Journal of Silk, 2017, 54(4): 43-51.

[3]陈敬玉, 张萌萌. 畲族彩带的要素特征及其在当代的嬗变[J]. 丝绸, 2018, 55(6): 83-90.

CHEN Jingyu, ZHANG Mengmeng. Element characteristics of colored sash of She nationality and its transmutation in contemporary era[J]. Journal of Silk, 2018, 55(6): 83-90.

[4]陈敬玉. 民族迁徙影响下的浙闽畲族服饰传承脉络[J]. 纺织学报, 2017, 38(4): 115-120.

CHEN Jingyu. Shes costumes heritage context in Zhejiang and Fujian provinces based on ethnic migration history[J]. Journal of Textile Research, 2017, 38(4): 115-120.

[5]吴微微, 陈良雨. 浙江畲族近代女子盛装审美艺术[J]. 纺织学报, 2008(1): 107-109.

WU Weiwei, CHEN Liangyu. Aesthetics and art of Shes modern women formal dress in Zhejiang[J]. Journal of Textile Research, 2008(1): 107-109.

[6]段婷. 畲族传统服饰面料及印染工艺分析[J]. 化纤与纺织技术, 2020, 49(10): 35-37.

DUAN Ting. Analysis of the traditional clothing fabrics and printing and dyeing process of She nationality[J]. Chemical Fiber & Textile Technology, 2020, 49(10): 35-37.

[7]徐平华, 冒海琳, 张玉杰, 等. 民族服饰色彩构成分析方法研究[J]. 丝绸, 2019, 56(3): 24-29.

XU Pinghua, MAO Hailin, ZHANG Yujie, et al. Study on analysis method for color composition of ethnic costumes[J]. Journal of Silk, 2019, 56(3): 24-29.

[8]HAGTVEDT H, BRASEL S A. Color saturation increases perceived product size[J]. Journal of Consumer Research, 2017, 44(2): 396-413.

[9]XU B Q, LIU X J, LU C F, et al. Transferring the color imagery from an image: A color network model for assisting color combination[J]. Color Research and Application, 2019, 44(2): 205-220.

[10]LIU X J, SUN Y, WU J F. Gene regulatory network model for product industrial design[J]. Journal of Theoretical and Applied Information Technology, 2013, 48(3): 1606-1611.

[11]徐明慧, 徐平華, 韦秋菊, 等. 基于网络关系模型的品牌女装配色解析[J]. 纺织学报, 2021, 42(11): 137-142.

XU Minghui, XU Pinghua, WEI Qiuju, et al. Color parsing of female brand clothing based on nexus network modeling[J]. Journal of Textile Research, 2021, 42(11): 137-142.

[12]夏帆. 畲族源生服装图系研究[M]. 杭州: 浙江人民美术出版社, 2017: 13-57.

XIA Fan. Research on Pattern of She Nationality Garment[M]. Hangzhou: Zhejiang Peoples Fine Arts Publishing House, 2017: 13-57.

[13]LEE G S, LEE S H, KIM G O, et al. A modified Grabcut using a clustering technique to reduce image noise[J]. Symmetry, 2016, 8(7): 64.

[14]BHATTACHARYA T, CHATTERJEE A. Evaluating performance of some common filtering techniques for removal of Gaussian noise in images[C]//2017 IEEE International Conference on Power, Control, Signals and Instrumentation Engineering (ICPCSI). Chennai: IEEE, 2017: 1981-1984.

[15]WANG D W, YAN W C, ZHU T, et al. An adaptive correction algorithm for non-uniform illumination panoramic images based on the improved bilateral gamma function[C]//2017 International Conference on Digital Image Computing: Techniques and Applications (DICTA). Sydney: IEEE, 2017: 1-6.

[16]LIU J, DU J, WANG X, et al. A hybrid algorithm based on improved LLE and k-means for visual codebook generation in scene classification[C]//2011 4th IEEE International Conference on Broadband Network and Multimedia Technology. Shenzhen: IEEE, 2011: 265-269.

[17]ANGELINE D M D. Association rule generation for student performance analysis using apriori algorithm[J]. The SIJ Transactions on Computer Science Engineering & Its Applications, 2013, 1(1): 12-16.

[18]ZHOU S C. Data mining and analysis of the compatibility law of traditional Chinese medicines based on FP-growth algorithm[J]. Journal of Mathematics, 2021(10): 1-10.

[19]WANG W. Optimization of intelligent data mining technology in big data environment[J]. Journal of Advanced Computational Intelligence and Intelligent Informatics, 2019, 23(1): 129-133.

[20]LA P T, LE B, VO B. Incrementally building frequent closed itemset lattice[J]. Expert Systems with Applications, 2014, 41(6): 2703-2712.

[21]李方园, 陈敬玉. 艺术人类学视角下的畲族服饰图案解读[J]. 包装工程, 2019, 40(2): 109-113.

LI Fangyuan, CHEN Jingyu. Interpretation of She costume patterns from the perspective of artistic anthropology[J]. Packaging Engineering, 2019, 40(2): 109-113.

[22]徐雯琦, 吳莹洁, 赵伟楠, 等. 族群认同影响下的贵州畲族服饰发展变迁[J]. 服装学报, 2020, 5(6): 520-526.

XU Wenqi, WU Yingjie, ZHAO Weinan. Development and changes of the She costume in Guizhou under the influence of ethnic identity[J]. Journal of Clothing Research, 2020, 5(6): 520-526.

Coloration association rules parsing of She nationality costumes

CAO Jingwena, JIA Jinga, XU Pinghuaa,b,c, LIN Ruibinga, SUN Xiaowana

(a.School of Fashion Design & Engineering; b.Zhejiang Provincial Research Center of Fashion Engineering Technology; c.MOC Key Laboratoryof Silk Culture Heritage and Product Design Digital Technology, Zhejiang Sci-Tech University, Hangzhou 310018, China)

Abstract:

Homogenization of textiles and garments, the mismatch between supply and demand, and slow design decisions have long been a bottleneck in the development of a high-quality industry. With the activation of consumer trends such as “oriental aesthetics” and “new fashion”, the regeneration of traditional, national and localized clothing spirit and cultural connotation has opened up new ideas and new ways for consumption growth. Color research based on image analysis technology is helpful to accurately, conveniently, and objectively characterize garment composition forms and color usage patterns and build a bridge between subjective perception and quantitative analysis, thus helping the development and application of intelligent color design for fashion products.

In order to clearly explain the color distribution and association rules of She traditional costumes, image analysis techniques were utilized to parse the imagery coloration relationship. Taking the She diaspora in Zhejiang, Jiangxi and Fujian provinces as an example, the coloration of 150 representative costumes images obtained from the field survey were analyzed. Firstly, the selected samples were subjected to image pre-processing operations. Secondly, in the construction stage of the base dataset, if conventional K-means clustering was used, the number of color clusters of each sample was forced to be uniform, which would easily lead to color lifting bias. Therefore, an improved dichotomy K-means adaptive clustering algorithm was used here to adaptively extract the color of each garment. Under HSV color space, the main colors of the sequence samples were extracted. On this basis, the K-means algorithm was used for secondary clustering when regional differences were compared horizontally, and the number of common color categories of She was determined according to relevant literature studies to unify the number of cluster centers and obtain the clothing imagery colors of each region. The improved vector set-based Apriori algorithm was used to resolve the multivariate color matching relationships among the imagery colors of She clothing, and to visually characterize the color patterns and correlations of different regional settings at the same time. Experimental results show that the quantified colors of She clothing in the three regions match the colors used in She clothing recorded in current literature, and the color palette as a whole presents black, red and blue. In the binary color matching relationship, there are some differences in the color matching color groups of She clothing in each region. In Zhejiang province, there are different shades of similar colors interwoven with each other, showing a rich sense of hierarchy; in Jiangxi province, the colors are deep and simple, with matching frequency of black and gray being greater than 50%, while in Fujian province, there are a small amount of red, yellow and other colors embellishing with each other. In the ternary color scheme relationship, the She clothing pairing is richer in Jiangxi province, simpler in Zhejiang province, and as a whole shows black, green, dark yellow and light yellow pairing relationships with each other in Fujian province. The average time cost of color matching rules parsing with the improved algorithm is 0.032 seconds, which can quickly parse the color correlation rules of She costumes coloration, and provides a referenced method of color analysis and regeneration design for other similar traditional costumes.

This study analyzes the color relationships of She clothing in Zhejiang, Jiangxi, and Fujian, and concretely represents the color usage patterns and color matching logic of clothing, which helps to realize the digital conservation of She clothing coloration. In addition, the objective and visualized way of characterizing the color mechanism of clothing provides a basis for color assignment for the future application of color activation and regeneration design of She clothing and further provides a methodological reference for the systematic study of similar traditional clothing colors.

Key words:

coloration relationship; association rules; adaptive clustering; color parsing; color spectrum; visualization

收稿日期:

2022-07-04;

修回日期:

2023-02-26

基金項目:

国家自然科学基金青年基金项目(61702460);国家社会科学基金重点项目(19AMZ009);浙江省高校重大人文社会科学攻关计划项目(2023QN092);浙江理工大学科研业务费专项资金资助项目(22076215-Y,2021Q057);服装设计国家级虚拟仿真实验教学中心项目(zx20212004);浙江省服装工程技术研究中心开放基金项目(2021FZKF05);浙江省教育厅科研基金项目(Y202250618);浙江理工大学教育教学改革研究重点项目(jgzd202202);浙江理工大学优秀研究生学位论文培育基金项目(LW-YP2021053)

作者简介:

曹竟文(1998),女,硕士研究生,研究方向为服饰色彩智慧设计。通信作者:徐平华,副教授,shutexph@163.com。

猜你喜欢

关联规则可视化
自然资源可视化决策系统
思维可视化
基于Power BI的油田注水运行动态分析与可视化展示
自然资源可视化决策系统
基于CGAL和OpenGL的海底地形三维可视化
“融评”:党媒评论的可视化创新
基于Apriori算法的高校学生成绩数据关联规则挖掘分析
基于关联规则和时间阈值算法的5G基站部署研究
关联规则,数据分析的一把利器
关联规则挖掘Apriori算法的一种改进