Scholarlink

<p>Dictionary indexing (DI) for Electron Backscatter Diffraction (EBSD) has conventionally traded computational speed for superior robustness to pattern noise when compared to traditional Hough indexing methods. To alleviate this computational bottleneck, we propose a novel Principal Component Analysis based Dictionary Indexing (PCA-DI) approach that uses dimensionality reduction and numerical precision optimization for accelerated pattern matching. A combination of synthetic benchmarks and benchmarks on a standard nickel dataset with varying noise levels demonstrate significant speedup factors over conventional DI that scale with image resolution while maintaining or improving indexing accuracy, particularly under high-noise conditions where dimensionality reduction suppresses noise-dominated high-frequency components. Implementation with reduced numerical precision (FP16 and dynamic INT8 quantization) provides additional performance improvements without compromising indexing quality. The resulting approach enables substantially faster dictionary indexing on consumer-grade hardware, addressing scalability challenges for modern high-resolution EBSD systems and low-symmetry materials.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Accelerating dictionary indexing of electron backscatter diffraction patterns with PCA and quantization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月4日

查看全文
  • Zachary T. Varley,
  • Gregory S. Rohrer,
  • Marc De Graef

摘要

Dictionary indexing (DI) for Electron Backscatter Diffraction (EBSD) has conventionally traded computational speed for superior robustness to pattern noise when compared to traditional Hough indexing methods. To alleviate this computational bottleneck, we propose a novel Principal Component Analysis based Dictionary Indexing (PCA-DI) approach that uses dimensionality reduction and numerical precision optimization for accelerated pattern matching. A combination of synthetic benchmarks and benchmarks on a standard nickel dataset with varying noise levels demonstrate significant speedup factors over conventional DI that scale with image resolution while maintaining or improving indexing accuracy, particularly under high-noise conditions where dimensionality reduction suppresses noise-dominated high-frequency components. Implementation with reduced numerical precision (FP16 and dynamic INT8 quantization) provides additional performance improvements without compromising indexing quality. The resulting approach enables substantially faster dictionary indexing on consumer-grade hardware, addressing scalability challenges for modern high-resolution EBSD systems and low-symmetry materials.