Details
Originalsprache | Englisch |
---|---|
Titel des Sammelwerks | SPIE Future Sensing Technologies 2023 |
Herausgeber/-innen | Osamu Matoba, Joseph A. Shaw, Christopher R. Valenta |
Herausgeber (Verlag) | SPIE |
ISBN (elektronisch) | 9781510657229 |
Publikationsstatus | Veröffentlicht - 22 Mai 2023 |
Veranstaltung | SPIE future sensing Technologies - Yokohama, Japan Dauer: 18 Apr. 2023 → 21 Apr. 2023 |
Publikationsreihe
Name | Proceedings of SPIE - The International Society for Optical Engineering |
---|---|
Band | 12327 |
ISSN (Print) | 0277-786X |
ISSN (elektronisch) | 1996-756X |
Abstract
ASJC Scopus Sachgebiete
- Werkstoffwissenschaften (insg.)
- Elektronische, optische und magnetische Materialien
- Physik und Astronomie (insg.)
- Physik der kondensierten Materie
- Mathematik (insg.)
- Angewandte Mathematik
- Ingenieurwesen (insg.)
- Elektrotechnik und Elektronik
- Informatik (insg.)
- Angewandte Informatik
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
SPIE Future Sensing Technologies 2023. Hrsg. / Osamu Matoba; Joseph A. Shaw; Christopher R. Valenta. SPIE, 2023. 123270D (Proceedings of SPIE - The International Society for Optical Engineering; Band 12327).
Publikation: Beitrag in Buch/Bericht/Sammelwerk/Konferenzband › Aufsatz in Konferenzband › Forschung
}
TY - GEN
T1 - Noise-robust registration of microscopic height data using convolutional neural networks
AU - Siemens, Stefan
AU - Kästner, Markus
AU - Reithmeier, Eduard
PY - 2023/5/22
Y1 - 2023/5/22
N2 - In this work, a deep convolutional neural network is proposed to improve the registration of microtopographic data. For this purpose, different mechanical surfaces were optically measured using a confocal laser scanning microscope. A wide range of surfaces with different materials, processing methods, and topographic properties, such as isotropy and anisotropy or stochastic and deterministic features, are included. Training and testing datasets with known homographies are generated from these measurements by cropping a fixed and moving image patch from each topography and then randomly perturbing the latter. A pseudo-siamese network architecture based on the VGG Net is then used to predict these homographies. The network is trained with a supervised learning approach where the Euclidean distance between the predicted and the ground truth gives the loss function. The 4-point homography parameterization is used to improve the loss convergence. Furthermore, different amounts of image noise are added to enhance the prediction’s robustness and prevent overfitting. The effectiveness of the proposed method is evaluated through different experiments. First, the network performance is compared to intensity-based and feature-based conventional registration algorithms regarding the resulting error, the noise-robustness, and the processing speed. In addition, images from the Microsoft Common Objects in Context (COCO) dataset are used to verify the network’s generalization capability to new image types and contents. The results show that the learning-based approach offers much higher robustness regarding image noise and a much lower processing time. In contrast, conventional algorithms have a smaller registration error without image noise.
AB - In this work, a deep convolutional neural network is proposed to improve the registration of microtopographic data. For this purpose, different mechanical surfaces were optically measured using a confocal laser scanning microscope. A wide range of surfaces with different materials, processing methods, and topographic properties, such as isotropy and anisotropy or stochastic and deterministic features, are included. Training and testing datasets with known homographies are generated from these measurements by cropping a fixed and moving image patch from each topography and then randomly perturbing the latter. A pseudo-siamese network architecture based on the VGG Net is then used to predict these homographies. The network is trained with a supervised learning approach where the Euclidean distance between the predicted and the ground truth gives the loss function. The 4-point homography parameterization is used to improve the loss convergence. Furthermore, different amounts of image noise are added to enhance the prediction’s robustness and prevent overfitting. The effectiveness of the proposed method is evaluated through different experiments. First, the network performance is compared to intensity-based and feature-based conventional registration algorithms regarding the resulting error, the noise-robustness, and the processing speed. In addition, images from the Microsoft Common Objects in Context (COCO) dataset are used to verify the network’s generalization capability to new image types and contents. The results show that the learning-based approach offers much higher robustness regarding image noise and a much lower processing time. In contrast, conventional algorithms have a smaller registration error without image noise.
KW - Confocal Laser Scanning Microscopy
KW - Convolutional Neural Network
KW - Homography estimation
KW - Image registration
KW - Machine learning
KW - Surface metrology
UR - http://www.scopus.com/inward/record.url?scp=85173430098&partnerID=8YFLogxK
U2 - 10.1117/12.2644620
DO - 10.1117/12.2644620
M3 - Conference contribution
T3 - Proceedings of SPIE - The International Society for Optical Engineering
BT - SPIE Future Sensing Technologies 2023
A2 - Matoba, Osamu
A2 - Shaw, Joseph A.
A2 - Valenta, Christopher R.
PB - SPIE
T2 - SPIE future sensing Technologies
Y2 - 18 April 2023 through 21 April 2023
ER -