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

Area of Science:

• Materials Science
• Crystallography
• Computational Methods

Background:

• Electron Backscatter Diffraction (EBSD) is crucial for materials characterization.
• Conventional Dictionary Indexing (DI) offers robustness but suffers from slow computational speed.
• Hough indexing methods are faster but less robust to pattern noise.

Purpose of the Study:

• To develop a faster and more accurate Dictionary Indexing (DI) method for Electron Backscatter Diffraction (EBSD).
• To address the computational bottleneck in conventional DI methods.
• To improve EBSD analysis for high-resolution systems and challenging materials.

Main Methods:

• Proposed a novel Principal Component Analysis based Dictionary Indexing (PCA-DI) approach.
• Utilized dimensionality reduction for accelerated pattern matching.
• Optimized numerical precision (FP16 and INT8 quantization) for performance gains.

Main Results:

• Achieved significant speedup factors over conventional DI, scaling with image resolution.
• Maintained or improved indexing accuracy, especially under high-noise conditions.
• Demonstrated substantial performance improvements on consumer-grade hardware.

Conclusions:

• PCA-DI offers a computationally efficient alternative to conventional DI for EBSD.
• The method enhances EBSD scalability for high-resolution datasets and low-symmetry materials.
• Reduced numerical precision further boosts performance without compromising accuracy.