컴퓨터 비전을 기반으로 한 중국 소수민족 색채 및 문양 추출과 가상 의상디자인에서의 응용 연구 : 중국 야오족, 먀오족, 좡족을 사례로 = A Study on the Extraction of Colors and Patterns of Ethnic Minorities in China Based on Computer Vision and Its Application in Virtual Costume Design

다국어 초록 (Multilingual Abstract)

In an era of rapid globalization and digitalization, the preservation
and transmission of intangible cultural heritage face new challenges and
opportunities. As an essential component of Chinese ethnic culture,
traditional handicrafts of ethnic minorities not only embody the
historical memory and aesthetic characteristics of specific ethnic groups
but also bear profound cultural connotations and social significance.
How to innovate based on this heritage to breathe new life into
intangible cultural heritage in a digital and modern context has become
a pressing research topic.
In recent years, the rapid development of virtual fashion, digital
cultural content, and interactive entertainment has provided extensive
application opportunities for the digitalization of intangible cultural
heritage. In virtual gaming and digital entertainment, the growing
consumer demand for diverse and personalized cultural experiences has
not only created market space for the digital application of ethnic
minority elements but also brought unprecedented opportunities for
their commercialization and international dissemination. Therefore, how
to effectively integrate these traditional cultural elements into the
design and dissemination within virtual environments has emerged as
both a focus and a challenge in the digital cultural content field.
To address this, this study aims to explore the integration of
technology and art, leveraging virtual character costume design to
investigate the feasibility of applying intangible cultural heritage
resources in digital design. Utilizing computer vision and deep learning
technologies, the study takes Yao embroidery, Miao embroidery, Miao
batik, and Zhuang brocade as examples, developing automated schemes
for color and pattern extraction specific to each research object. The
extracted data undergoes color feature analysis, forming a color
network model and a pattern database to support subsequent design
applications. Additionally, to meet the demand for rapid customization of
color styles for virtual character costumes, a fast tone-switching
algorithm based on grayscale mapping has been developed, alongside a
global style transfer model that supports real-time conceptual design
for virtual characters, thus providing technical support for integrating
traditional patterns with modern design aesthetics.
This study adopts multiple research methodologies aligned with these
objectives, including literature review, field investigation, algorithm
testing and optimization, statistical analysis, mathematical modeling, and
deep learning neural network training. Initially, a literature review
established the theoretical foundation for digitalization in intangible
cultural heritage, computer vision, and deep learning, summarizing prior
research in these fields. During the data collection phase, field
investigations were conducted with on-site photography, supplemented
by public resources from local museums, resulting in a comprehensive
image dataset of the research objects. The experimental design,
implemented on the MATLAB platform, involved image preprocessing,
segmentation, clustering, and morphological operations to optimize the
extraction of color and pattern features. Throughout the experimental
process, algorithm testing and optimization were employed to compare,
fine-tune, and enhance the effectiveness of different parameter
outputs. Combined with image feature analysis, this optimization process
improved the reliability and applicability of color and pattern extraction
schemes. A color conversion algorithm was developed through
mathematical modeling, and transfer learning in deep learning was
employed to train neural networks and adjust model parameters,
thereby enhancing the stability and performance of the style transfer
model.
The experimental results demonstrate that:
For color extraction in Yao embroidery, bilateral filtering optimized
boundary features, and effective image segmentation was achieved
using Euclidean distance in the LAB color space, followed by K-means
clustering for color extraction. In pattern extraction, morphological
operations with connected region labeling enabled the effective
separation of independent pattern units.
For Miao embroidery, color clustering utilized bilateral filtering and
non-local means filtering to enhance edge detail, with superpixel
segmentation summarizing image features. DBSCAN clustering with
K-distance achieved effective color clustering and principal component
extraction. Gabor filtering was introduced in pattern extraction to
enhance texture features, and binarization with morphological operations
extracted pattern boundaries.
For Miao batik, median filtering smoothed the image, while gradient
vector flow and edge detection techniques precisely captured pattern
contours. The technique for Zhuang brocade employed Gaussian
filtering to enhance structural features, Fourier transform for periodic
analysis to identify repetitive structures, and feature point matching
with autocorrelation analysis to isolate repeating units, establishing the
foundational conditions for color and pattern extraction.
The color transformation algorithm developed from the extracted
color and pattern data preserved the original pattern structure and
enabled fast style-switching across images, providing a feasible
technical pathway for real-time customized tone conversion of ethnic
patterns in virtual costume design.
The VGG-19 pre-trained network, trained via transfer learning,
extracted global features of the style image through the GRAM feature
extraction layer, successfully integrating with custom patterns to support
the conceptual design needs of virtual character costumes.
Through this research, this study aims to provide technical support
for the application of Chinese ethnic minority cultural heritage within
the digital cultural and creative industries, combined with virtual
costume design, to meet contemporary user demand for diverse cultural
experiences and the global dissemination of local culture. The study
aspires to establish a feasible technical path for the development of
intangible cultural heritage in digitalization and commercialization.
This study, however, has certain limitations. At this stage, the
research focuses primarily on algorithmic applications and technical
development, lacking market feedback on the application of technical
schemes in virtual character costume design, and there remains room
for improvement in algorithmic efficiency and resource allocation.
Future research will combine deep learning models with high-efficiency
computing power to further enhance the applicability and scalability of
the algorithms. Additionally, future plans include expanding to 3D image
processing, exploring style transfer of Chinese ethnic art symbols in
three-dimensional space with virtual reality support through
multi-dimensional feature fusion.

국문 초록 (Abstract)

글로벌화와 디지털화의 급속한 발전 속에서 비물질 문화유산의 보호와 전
승은 새로운 도전과 기회를 맞이하고 있다. 중국 민족 문화의 중요한 구성 요소
인 소수 민족의 전통 수공예품은 특정 민족의 역사적 기억과 미적 특징을 집약
할 뿐만 아니라, 풍부한 문화적 함의와 사회적 의미를 내포하고 있다. 이러한
비물질 문화유산을 전승하는 동시에 디지털화와 현대화라는 맥락에서 새로운
생명력을 불어넣는 방안을 모색하는 것은 현재 시급히 연구가 요구되는 과제이
다.최근 들어, 가상 패션, 디지털 문화 콘텐츠, 인터랙티브 엔터테인먼트 분야가
빠르게 발전하면서 비물질 문화유산의 디지털화에 넓은 응용 가능성을 제공하
고 있다. 가상 게임과 디지털 엔터테인먼트 분야에서는 소비자의 다원적이고 개
인화된 문화 체험에 대한 요구가 날로 증가하고 있으며, 이러한 요구는 소수 민
족 전통 요소의 디지털화된 응용에 시장 공간을 제공할 뿐만 아니라, 상업화와
국제적 전파에도 전례 없는 기회를 가져왔다. 따라서 이러한 민족 전통 문화 요
소를 가상 환경의 디자인과 전파에 효과적으로 융합시키는 방안을 찾는 것이
디지털 문화 콘텐츠 분야의 주요한 관심사이자 난제로 떠오르고 있다. 이에 본 연구는 기술과 예술의 융합 연구에 초점을 맞추고, 가상 캐릭터 의상
디자인과 결합하여 비물질 문화유산 자원의 디지털 디자인 응용 가능성을 탐색- ii -
하고자 한다. 컴퓨터 비전과 딥러닝 기술을 바탕으로, 야오족((瑤族) 자수, 먀오
족(苗族) 자수, 먀오족 바틱, 좡족(壯族) 금을 예로 들어, 다양한 연구 대상에
적합한 색상 및 문양 자동화 추출 방안을 구축하고, 추출 결과에 대한 색상 특
성 분석을 통해 색상 네트워크 모델 및 문양 데이터베이스를 확립하여 후속 디
자인 응용에 기초 데이터를 제공한다. 또한, 가상 공간에서 빠르게 캐릭터 의상
의 색상 스타일을 맞춤형으로 설정할 수 있는 요구를 충족하기 위해, 본 연구는
그레이스케일 계층 맵핑을 기반으로 한 색조 빠른 전환 알고리즘을 개발하고, 실시간 개념 디자인을 지원하는 전역 스타일 전이 모델을 구축하여 전통 문양
과 현대 디자인 스타일의 융합을 위한 기술적 지원을 제공하였다. 본 연구는 위의 연구 내용에 따라 다양한 연구 방법을 채택하였으며, 문헌 리
뷰, 현장 조사, 알고리즘 테스트 및 최적화, 통계 분석, 수학적 모델링, 딥러닝
신경망 학습 등을 포함한다. 우선, 문헌 리뷰를 통해 본 연구가 비물질 문화유
산 디지털화, 컴퓨터 비전 및 딥러닝 분야에서 이론적 기초를 확립하고, 관련
분야의 선행 연구 성과를 정리하였다. 데이터 수집 단계에서는 현장 조사법을
사용하여 현지에서 촬영한 자료와 현지 박물관의 공개 자료를 결합하여 연구
대상의 이미지 데이터셋을 구축하였다. 실험 설계에서는 MATLAB 플랫폼을 활용하여 이미지 전처리, 분할, 군집화
및 형태학적 연산 실험을 진행하여 색상과 문양 특성의 추출 효과를 최적화하
였다. 실험 과정에서 알고리즘 테스트와 최적화를 통해 서로 다른 알고리즘 매
개변수의 출력 결과를 비교하고, 조정 및 개선하며 이미지 특성 분석을 결합하
여 각 실험 단계의 알고리즘 흐름을 최적화함으로써 색상과 문양 추출 방안의
신뢰성과 적합성을 향상시켰다. 수학적 모델링을 기반으로 문양 색상 변환 알고
리즘을 개발하고, 딥러닝의 전이 학습을 결합하여 신경망 학습과 모델 매개변수
조정을 통해 스타일 전이 모델의 안정성과 성능을 강화하였다. 실험 결과는 다음과 같다:
첫째, 야오족 자수의 색상 추출에서는 양측 필터를 사용하여 경계 특성을 최
적화하고, LAB 색상 공간의 유클리드 거리 방법을 통해 효과적인 이미지 분할
을 구현하였으며, K-means 색상 군집화를 결합하여 색상 추출을 완료하였다. 문양 추출에서는 형태학적 연산의 연결 영역 레이블링을 결합하여 독립된 문양
요소를 효과적으로 분리할 수 있었다. 둘째, 먀오족 자수의 색상 군집화에서는 양측 필터와 비국소적 평균 필터를 - iii -
통해 문양의 경계 세부 사항을 최적화하고, 초픽셀 분할을 이용하여 이미지 특
성을 개괄한 후, DBSCAN 군집화 알고리즘과 K-거리 방법을 결합하여 색상의
효과적인 군집화와 주성분 추출을 달성하였다. 문양 추출에서는 Gabor 필터를
도입하여 텍스처 특성을 강화하고, 이진화 분할과 형태학적 연산을 통해 문양
경계를 추출하였다. 셋째, 먀오족 바틱 문양 추출에서는 중간 필터를 이용하여 이미지를 평활화하
고, 그래디언트 벡터 흐름과 경계 검출 기술을 결합하여 문양의 윤곽을 정확하
게 추출하였다. 좡족 금의 기술 방안은 가우시안 필터를 통해 구조 특성을 부각
하고, 푸리에 변환의 주기 분석을 통해 주기적 구조를 인식하며, 특징점 매칭과
자기 상관분석을 통해 반복 단위 구조를 분리하여 색상 및 문양 추출의 기초
조건을 구축하였다. 넷째, 색상 및 문양 추출 데이터를 기반으로 개발된 색상 변환 알고리즘은 원
래 문양 구조를 유지하면서 이미지 간의 빠른 색상 스타일 전환을 실현하여 가
상 의상디자인에서 민족 문양의 실시간 맞춤형 색조 변환을 위한 기술 경로를
제공하였다. 다섯째, 전이 학습을 통해 학습된 VGG-19 사전 학습 네트워크는 GRAM 특성
추출 층을 기반으로 스타일 이미지의 전역 특성 추출을 구현하였으며, 사용자
지정 문양과의 스타일 융합을 성공적으로 달성하여 가상 캐릭터 의상디자인 수
요에 기술 지원을 제공하였다. 이상의 연구를 통해 본 연구는 중국 소수 민족 문화유산이 디지털 문화 창의
산업에서 응용될 수 있도록 기술적 지원을 제공하며, 가상 의상 디자인과 결합
하여 현대 소비자의 다원적 문화 체험 요구와 중국 전통문화의 글로벌 확산에
정보제공과 더불어 중국 소수 민족 문화의 현대적 전승을 위한 촉진역할 그리
고 비물질 문화유산이 디지털화 및 상업화 방향으로 발전하는 데 있어 실현 가
능한 기술 경로를 제시하는데 목적을 두고 있다. .

목차 (Table of Contents)

• I. 서 론
• 1. 연구배경 및 연구목적 ·································································· 1
• 가. 중국 소수민족의 무형 문화유산 연구 배경 ································· 1
• 나. 시대적 배경 및 산업적 배경 ························································· 3
• 다. 연구 목적 ························································································ 4
• 2. 연구 방법 및 연구 내용 ····························································· 6
• 가. 연구 방법 ························································································ 6
• 나. 연구 내용 ························································································ 7
• 3. 연구흐름도 ······················································································· 8
• Ⅱ. 이론적 배경
• 1. 중국 소수민족 전통 자수와 바틱 기술 개요 ······················ 9
• 가. 야오족 자수 ···················································································· 9
• 나. 먀오족 자수와 먀오족 바틱 ························································· 10
• 다. 좡족 전통 금직물 ········································································· 12
• 2. 이미지 전처리의 이론 ································································ 13
• 가. 양방향 필터링 기술 원리 ···························································· 15
• 나. 비국소적 평균 필터링 기술 원리 ················································ 16
• 다. 중간값 필터링 기술 원리 ···························································· 17
• 라. 가버 필터링 기술 원리 ································································ 18
• 마. 이미지 전처리 이론 소결 ···························································· 19
• 3. 이미지 분할 이론 ········································································22
• 가. 색채 기반 분할 ············································································· 23
• 나. 임계값 이진화를 활용한 분할 ····················································· 24
• 다. 영역 기반 분할 ············································································· 26
• 라. 특징 기반 분할 ············································································· 27
• 마. 이미지 분할 이론 요약 ································································ 29
• 4. 색채 클러스터링 이론 ································································ 31
• 가. K-means 색채 클러스터링 ··························································· 33
• 나. DBSCAN 색채 클러스터링 ··························································· 34
• 5. 형태학적 작업 이론 ···································································· 37
• 6. 문양 및 구조 분석 이론 ··························································· 39
• 가. 푸리에 변환과 자기상관분석 ······················································· 39
• 나. 에지 검출 ···················································································· 41
• 다. 경계 벡터 흐름(GVF) ·································································· 44
• 7. 스타일 전이 이론적 배경 ························································· 47
• 가. 관련 기술 개요 ············································································· 47
• 나. 주요 기술 방법 ············································································· 47
• 1) 합성곱 신경망(CNN) ····································································· 47
• 2) 생성적 적대 신경망(GAN) ···························································· 48
• Ⅲ. 기술 경로 및 기술 흐름
• 1. 기술 경로 흐름도 ······································································ 50
• 2. 기술 프로세스 개요 ···································································· 51
• 가. 야오족 자수 색채 및 문양 추출 ················································· 51
• 나. 먀오족 자수 색채 및 문양 추출 ················································· 52
• 다. 좡족 자수 색채 및 문양 추출 ·····················································52
• Ⅳ. 색채 및 문양 추출 실험
• 1. 야오족 자수 색채 및 문양 추출 ············································ 54
• 가. 야오족 자수 이미지 특징 분석 ··················································· 54
• 나. 야오족 자수 색채 및 문양 추출 절차와 실험 단계 ·················· 55
• 다. K-means 알고리즘을 기반으로 한 야오족 자수 색채 추출 ······ 56
• 1) 야오족 자수 이미지 필터링 ························································· 56
• 2) LAB색채 공간 변환 및 이미지 분할 ··········································· 58
• 3) K-means 색채 클러스터링과 클러스터링 결과 분석 ················ 60
• 4) 야오족 자수 색채 추출 기술의 알고리즘 요약 ························· 66
• 라. 형태학 및 연결 영역 레이블링을 통한 문양 추출 ···················· 66
• 1) LAB 색채 공간 기반 문양 분할 ·················································· 66
• 2) 이미지 이진화 및 형태학적 개방 연산 ······································ 68
• 3) 연결 영역 레이블링과 목표 탐지 ··············································· 70
• 2. 먀오족 자수와 바틱의 색채 및 문양 추출 ························ 75
• 가. 먀오족 자수 이미지 특징 분석 ··················································· 75
• 나. 먀오족 자수 색채 추출 연구 프레임워크 ··································· 76
• 다. 초픽셀 분할 기반 DBCSAN 색채 클러스터링 ···························· 77
• 1) 먀오족 자수 이미지 필터링 ························································· 77
• 2) 먀오족 자수 초픽셀 분할 ···························································· 79
• 3) DBSCAN 색채 클러스터링 실험 데이터 분석 ···························· 82
• 4) 먀오족 자수 색채 구성 분석 ······················································· 85
• 5) 먀오족 자수 색채 모델 구축 ······················································· 87
• 6) 먀오족 자수 색채 추출 기술 요약 ·············································· 89
• 라. 먀오족 납염 이미지 특징 분석 ··················································· 90
• 마. 먀오족 바틱 문양 추출 기술 프레임워크 ··································· 92
• 바. GVF 기반 먀오족 바틱 가장자리 추출 ·······································93
• 1) 이미지 전처리 ··············································································· 93
• 2) 이진화 문양 분할 ········································································· 94
• 3) 그레이디언트 벡터 필드(GVF)와 가장자리 검출 ······················· 94
• 4) GVF 기반 가장자리 추출 ···························································· 95
• 5) Sobel 연산자와 가장자리 세분화 ················································ 96
• 사. 먀오족 자수 문양 추출 ································································ 96
• 1) 먀오족 자수 이미지의 주파수 특성 분석 ··································· 96
• 2) 먀오족 자수 이미지 Gabor 필터링 ············································· 97
• 3) 이진화 이미지 분할 및 형태학적 처리 ······································ 98
• 4) 먀오족 자수 가장자리 추출 ························································· 99
• 5) 문양 추출 실험 결과 ·································································· 100
• 6) 먀오족 자수 및 바틱 문양 추출 기술 요약 ····························· 101
• 3. 좡족 장금 색채 및 문양 추출 ·············································· 103
• 가. 장금 이미지 특성 분석 ······························································· 103
• 나. 장금 색채 및 문양 추출 기술 프레임워크 ································ 105
• 다. 이미지 전처리 ············································································· 106
• 라. 푸리에 주기성 분석 ···································································· 108
• 마. 특징 매칭 및 자기 상관분석 ······················································ 110
• 바. 특징 군집화 및 형태학적 분할 ·················································· 112
• 사. 색채 및 문양 추출 ······································································ 114
• 아. 장금 색채 및 문양 추출 기술 프레임워크 요약 ······················ 116
• Ⅴ. 디자인 응용
• 1. 문양 호출 ····················································································· 118
• 2. 색채 스타일 변환 ······································································ 121
• 가. 기술 원리와 모델링 ····································································122
• 나. 알고리즘 적용 ············································································· 123
• 3. 향상된 VGG-19 네트워크 기반 스타일 전이 ·················· 126
• 가. 네트워크 구조와 특징 추출 ······················································· 126
• 나. 손실 함수 ···················································································· 127
• 다. 최적화 및 훈련 ··········································································· 128
• 4. 디자인 응용 요약 ······································································ 131
• Ⅵ. 결 론
• 1. 연구의 요약 및 연구시사점 ················································· 132
• 가. 야오족 색채와 문양 추출 ································································ 132
• 나. 먀오족 색채 및 문양 추출 ······························································ 133
• 다. 좡족 색채 및 문양 추출 ·································································· 135
• 라. 색채 스타일 변환 알고리즘 개발 및 응용 ·································· 136
• 마. 딥러닝 기반 스타일 전이 신경망 모델 개발 ······························ 137
• 2. 연구의 한계와 적용 가능성 ··················································· 138
• 가. 연구의 한계 ························································································ 138
• 나. 연구의 적용 가능성 ·········································································· 138
• 참고문헌
• Abstract 중국어 요약