| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
An Edgeworth Cross Mutual Information Function for Multimodal Pattern Matching |
| Author(s) |
Zachary Varley, Megna Shah, Jeff Simmons |
| On-Site Speaker (Planned) |
Zachary Varley |
| Abstract Scope |
In many materials engineering contexts, especially serial sectioning experiments, disparate micrographs from distinct sensors need to be registered to provide a more holistic perspective of a sample. Often, conventional computer vision techniques struggle to perform in these multimodal problems. One tool, mutual information can be used to rectify these situations, but it can be computationally intensive to compute between two images over groups such as image translation or rigid motion. We cover existing FFT-based methods for computing mutual information over such spaces, and present a novel extension that uses Edgeworth series expansions to dramatically accelerate this task. Specifically, we present 3rd and 4th order Edgeworth series estimates of mutual information and demonstrate these approaches on example multimodal data registration problems in the scanning electron microscope. We compare the runtime and accuracy trade-off with existing cross mutual information algorithms. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, |