서브메뉴
검색
Data-Driven Techniques for Structural Texture Analysis
Data-Driven Techniques for Structural Texture Analysis
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211153043
- ISBN
- 9798346857549
- DDC
- 621.3
- 저자명
- Zhang, Kaixuan.
- 서명/저자
- Data-Driven Techniques for Structural Texture Analysis
- 발행사항
- [Sl] : Northwestern University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 110 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-06, Section: B.
- 주기사항
- Advisor: Pappas, Thrasyvoulos.
- 학위논문주기
- Thesis (Ph.D.)--Northwestern University, 2024.
- 초록/해제
- 요약Texture is a critical component of image processing and computer vision, as it provides essential information for object and scene understanding. The main thrusts of this thesis are on data-driven techniques for optimizing the performance of structural texture similarity metrics (STSIMs) in specific application domains, and the use of statistical texture features for material characterization in medical imaging and other fields.We propose methods for enhancing STSIMs by learning the relationship between subband statistics and human perception in the context of specific application domains. Our approach leverages gradient descent algorithms to optimize STSIMs, allowing for a more accurate alignment with human judgments. The effectiveness of these techniques depends on systematically constructed annotated databases. These include identical textures, distorted textures, nonuniform textures, and generative neural network-synthesized textures. Experimental results demonstrate that data-driven STSIMs, not only improve texture similarity assessment, but also perform competitively with state-of-the-art convolutional neural network-based metrics at a lower computational cost.This thesis also explores the application of STSIM metrics and statistical texture features in film grain quality evaluation for video compression and medical image analysis. We show that traditional metrics like MSE, SSIM, and VMAF do not adequately reflect human perception of film grain, whereas our trained STSIMs offer more accurate evaluations. We also developed a real-time film grain analysis and synthesis algorithm based on a physical Boolean model. In the medical domain, we applied STSIM-inspired statistical texture features to chest X-Ray analysis and brain MRI analysis, demonstrating their ability to predict conditions such as bronchopulmonary dysplasia (BPD) and Cerebral Small Vessel Disease (CSVD). The findings underscore the potential of texture analysis in detecting subtle microstructural changes in the brain, contributing to a better understanding of how these changes correlate with conditions like hypertension and cognitive decline.Overall, this thesis demonstrates the versatility and effectiveness of the proposed texture similarity metrics across different domains, paving the way for further advancements in both theoretical and applied aspects of texture analysis.
- 일반주제명
- Electrical engineering
- 일반주제명
- Computer engineering
- 키워드
- Texture
- 키워드
- Image processing
- 기타저자
- Northwestern University Electrical and Computer Engineering
- 기본자료저록
- Dissertations Abstracts International. 86-06B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017164766
■00520250211153043
■006m o d
■007cr#unu||||||||
■020 ▼a9798346857549
■035 ▼a(MiAaPQ)AAI31639507
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a621.3
■1001 ▼aZhang, Kaixuan.
■24510▼aData-Driven Techniques for Structural Texture Analysis
■260 ▼a[Sl]▼bNorthwestern University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a110 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-06, Section: B.
■500 ▼aAdvisor: Pappas, Thrasyvoulos.
■5021 ▼aThesis (Ph.D.)--Northwestern University, 2024.
■520 ▼aTexture is a critical component of image processing and computer vision, as it provides essential information for object and scene understanding. The main thrusts of this thesis are on data-driven techniques for optimizing the performance of structural texture similarity metrics (STSIMs) in specific application domains, and the use of statistical texture features for material characterization in medical imaging and other fields.We propose methods for enhancing STSIMs by learning the relationship between subband statistics and human perception in the context of specific application domains. Our approach leverages gradient descent algorithms to optimize STSIMs, allowing for a more accurate alignment with human judgments. The effectiveness of these techniques depends on systematically constructed annotated databases. These include identical textures, distorted textures, nonuniform textures, and generative neural network-synthesized textures. Experimental results demonstrate that data-driven STSIMs, not only improve texture similarity assessment, but also perform competitively with state-of-the-art convolutional neural network-based metrics at a lower computational cost.This thesis also explores the application of STSIM metrics and statistical texture features in film grain quality evaluation for video compression and medical image analysis. We show that traditional metrics like MSE, SSIM, and VMAF do not adequately reflect human perception of film grain, whereas our trained STSIMs offer more accurate evaluations. We also developed a real-time film grain analysis and synthesis algorithm based on a physical Boolean model. In the medical domain, we applied STSIM-inspired statistical texture features to chest X-Ray analysis and brain MRI analysis, demonstrating their ability to predict conditions such as bronchopulmonary dysplasia (BPD) and Cerebral Small Vessel Disease (CSVD). The findings underscore the potential of texture analysis in detecting subtle microstructural changes in the brain, contributing to a better understanding of how these changes correlate with conditions like hypertension and cognitive decline.Overall, this thesis demonstrates the versatility and effectiveness of the proposed texture similarity metrics across different domains, paving the way for further advancements in both theoretical and applied aspects of texture analysis.
■590 ▼aSchool code: 0163.
■650 4▼aElectrical engineering
■650 4▼aComputer engineering
■653 ▼aTexture
■653 ▼aImage processing
■653 ▼aSubband statistics
■690 ▼a0544
■690 ▼a0464
■71020▼aNorthwestern University▼bElectrical and Computer Engineering.
■7730 ▼tDissertations Abstracts International▼g86-06B.
■790 ▼a0163
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17164766▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


