Andreas Uhl
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Andreas Uhl has a total of 218 co-authors. Here are the top co-authors by number of shared publications:
- Christof Kauba
University of Salzburg
17 shared publications - Georg Wimmer
Department of Artificial Intelligence and Human Interfaces Paris Lodron University Salzburg Salzburg Austria
15 shared publications - Heinz Hofbauer
Paris-Lodron-University Salzburg
14 shared publications - Michael Liedlgruber
University of Salzburg
9 shared publications - Robert Jöchl
Department of Artificial Intelligence and Human Interfaces University of Salzburg Salzburg Austria
8 shared publications - Simon Kirchgasser
University of Salzburg, Salzburg, Austria
8 shared publications
Latest Publications
- 28 September 2026
- Published by Springer Nature
- p. 49-62
Improving Smartphone-Based Dental Ceramic Manufacturer Classification Across Different Devices
- 28 September 2026
- Published by Springer Nature
- p. 49-62
Improving Smartphone-Based Dental Ceramic Manufacturer Classification Across Different Devices
- 28 September 2026
- Published by Springer Nature
- p. 558-570
Quantifying the Impact of Segmentation Inaccuracies on Cell Phenotyping in Imaging Mass Cytometry
- 28 September 2026
- Published by Springer Nature
- p. 558-570
Quantifying the Impact of Segmentation Inaccuracies on Cell Phenotyping in Imaging Mass Cytometry
- Open Access
- 16 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 64-74
Are Steganalysis Models Suitable for Temporal Image Forensics?
- 16 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 64-74
Open AccessAre Steganalysis Models Suitable for Temporal Image Forensics?
- 12 March 2026
- Published by Springer Nature
- p. 299-306
Medical Image Annotations for AI-based Wound Segmentation
- 12 March 2026
- Published by Springer Nature
- p. 299-306
Medical Image Annotations for AI-based Wound Segmentation
- 13 January 2026
- Published by Springer Nature
- p. 246-258
Wenceslas or Not - Face Comparison on Frana’s Wenceslas Depiction in the Wenceslas Bible
- 13 January 2026
- Published by Springer Nature
- p. 246-258
Wenceslas or Not - Face Comparison on Frana’s Wenceslas Depiction in the Wenceslas Bible
- 2 January 2026
- Published by Springer Nature
- p. 499-510
Face Detection in Historic Manuscripts A Case Study on the Wenceslas Bible
- 2 January 2026
- Published by Springer Nature
- p. 499-510
Face Detection in Historic Manuscripts A Case Study on the Wenceslas Bible
- Open Access
- 1 January 2026
- Published by Institution of Engineering and Technology (IET) in IET Biometrics
- Vol. 2026 (1) , 5158377
Stability of EEG‐Biometrics: The Role of Artifacts in Within vs. Cross Session Authentication
Artifacts in the electroencephalogram (EEG) have not been investigated systematically regarding their effect on EEG biometric performance. Since artifacts vary over time, they can be expected to lower EEG biometric performance in cross session cross‐validation. However, because of their presumed uniqueness, they might increase within session cross‐validation performance of an EEG biometric system. We examined the EEG biometric performance of 15 features within and across 6 sessions recorded over 3 days, including 3 different cognitive conditions, before and after artifact exclusion in a sample of up to 90 individuals. Our results revealed significant effects of artifact exclusion depending on the cross‐validation scenario and the feature. Overall, within session cross‐validation was not systematically affected differently by artifact exclusion as compared to cross session cross‐validation, but depending on the feature and the type of cognitive task employed, artifact exclusion led to better or worse results, with some frequency‐dependent features performing better when artifacts were excluded. Our results emphasize the overestimation of biometric performance in within session cross‐validation (with equal error rates (EERs) down to 0) as compared to cross session cross‐validation (with EERs for the same feature above 30) and the low reliability of results obtained when evaluating EEG biometric performance in small samples. We recommend that future developments of EEG biometrics must test their robustness against artifacts in the data.
- 1 January 2026
- Published by Institution of Engineering and Technology (IET) in IET Biometrics
- Vol. 2026 (1) , 5158377
Open AccessStability of EEG‐Biometrics: The Role of Artifacts in Within vs. Cross Session Authentication
Artifacts in the electroencephalogram (EEG) have not been investigated systematically regarding their effect on EEG biometric performance. Since artifacts vary over time, they can be expected to lower EEG biometric performance in cross session cross‐validation. However, because of their presumed uniqueness, they might increase within session cross‐validation performance of an EEG biometric system. We examined the EEG biometric performance of 15 features within and across 6 sessions recorded over 3 days, including 3 different cognitive conditions, before and after artifact exclusion in a sample of up to 90 individuals. Our results revealed significant effects of artifact exclusion depending on the cross‐validation scenario and the feature. Overall, within session cross‐validation was not systematically affected differently by artifact exclusion as compared to cross session cross‐validation, but depending on the feature and the type of cognitive task employed, artifact exclusion led to better or worse results, with some frequency‐dependent features performing better when artifacts were excluded. Our results emphasize the overestimation of biometric performance in within session cross‐validation (with equal error rates (EERs) down to 0) as compared to cross session cross‐validation (with EERs for the same feature above 30) and the low reliability of results obtained when evaluating EEG biometric performance in small samples. We recommend that future developments of EEG biometrics must test their robustness against artifacts in the data.
- Open Access
- 16 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 180344-180354
A Benchmark of Neural Networks for Semantic Segmentation of Wood Log Ends
Neural networks have achieved significant performance in image segmentation tasks. This paper describes a benchmark of deep learning-based methods for segmenting log ends of several species in images. More precisely, we are interested in segmenting a cross-section of a log positioned approximately at the centre of the image. Other logs that may be present next to the log of interest in the image should not be segmented. This task of semantic segmentation is essential in various applications, such as biometric log tracking and log quality assessment. Different state-of-the-art Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) for segmentation were tested in this paper, including Mask R-CNN, U-Net, PointRend, YOLOv11 and OneFormer. Some of these, being among the most recent architectures, had never before been tested on this type of image. To compare efficiency of the networks, well-known metrics such as Precision, Recall, Accuracy, Dice, IoU and MCC were used. From the experiment results, PointRend found to produce the most accurate segmentations, while YOLOv11 was better suited to real-time processing. In addition, an original dataset consisting of 1345 manually segmented images of oak log ends was created for this study and made available. Oak is a new species in these segmentation approaches. This dataset was supplemented with 855 images of two softwood species (Douglas fir and spruce) in order to increase the genericity of the models. The images have been selected to represent real situations encountered in the field and a wide variety of acquisition conditions and cross-section appearance. The code is available here : https://github.com/Braquemarok/Log_segmentation_benchmark.git.
- 16 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 180344-180354
Open AccessA Benchmark of Neural Networks for Semantic Segmentation of Wood Log Ends
Neural networks have achieved significant performance in image segmentation tasks. This paper describes a benchmark of deep learning-based methods for segmenting log ends of several species in images. More precisely, we are interested in segmenting a cross-section of a log positioned approximately at the centre of the image. Other logs that may be present next to the log of interest in the image should not be segmented. This task of semantic segmentation is essential in various applications, such as biometric log tracking and log quality assessment. Different state-of-the-art Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) for segmentation were tested in this paper, including Mask R-CNN, U-Net, PointRend, YOLOv11 and OneFormer. Some of these, being among the most recent architectures, had never before been tested on this type of image. To compare efficiency of the networks, well-known metrics such as Precision, Recall, Accuracy, Dice, IoU and MCC were used. From the experiment results, PointRend found to produce the most accurate segmentations, while YOLOv11 was better suited to real-time processing. In addition, an original dataset consisting of 1345 manually segmented images of oak log ends was created for this study and made available. Oak is a new species in these segmentation approaches. This dataset was supplemented with 855 images of two softwood species (Douglas fir and spruce) in order to increase the genericity of the models. The images have been selected to represent real situations encountered in the field and a wide variety of acquisition conditions and cross-section appearance. The code is available here : https://github.com/Braquemarok/Log_segmentation_benchmark.git.
- Open Access
Temporal Image Forensics: A Review and Critical Evaluation
Temporal image forensics is the science of estimating the age of a digital image. Usually, time-dependent traces (age traces) introduced by the image acquisition pipeline are exploited for this purpose. In this review, a comprehensive overview of the field of temporal image forensics based on time-dependent traces from the image acquisition pipeline is given. This includes a detailed insight into the properties of known age traces (i.e., in-field sensor defects and sensor dust) and temporal image forensics techniques. Another key aspect of this work is to highlight the problem of content bias and to illustrate how important eXplainable Artificial Intelligence methods are to verify the reliability of temporal image forensics techniques. Apart from reviewing material presented in previous works, in this review: (i) a new (probably more realistic) forensic setting is proposed; (ii) the main properties (growth rate and spatial distribution) of in-field sensor defects are verified; (iii) it is shown that a method proposed to utilize in-field sensor defects for image age approximation actually exploits other traces (most likely content bias); (iv) the features learned by a neural network dating palmprint images are further investigated; (v) it is shown how easily a neural network can be distracted from learning age traces. For this purpose, previous work is analyzed, re-implemented if required and experiments are conducted.
Open AccessTemporal Image Forensics: A Review and Critical Evaluation
Temporal image forensics is the science of estimating the age of a digital image. Usually, time-dependent traces (age traces) introduced by the image acquisition pipeline are exploited for this purpose. In this review, a comprehensive overview of the field of temporal image forensics based on time-dependent traces from the image acquisition pipeline is given. This includes a detailed insight into the properties of known age traces (i.e., in-field sensor defects and sensor dust) and temporal image forensics techniques. Another key aspect of this work is to highlight the problem of content bias and to illustrate how important eXplainable Artificial Intelligence methods are to verify the reliability of temporal image forensics techniques. Apart from reviewing material presented in previous works, in this review: (i) a new (probably more realistic) forensic setting is proposed; (ii) the main properties (growth rate and spatial distribution) of in-field sensor defects are verified; (iii) it is shown that a method proposed to utilize in-field sensor defects for image age approximation actually exploits other traces (most likely content bias); (iv) the features learned by a neural network dating palmprint images are further investigated; (v) it is shown how easily a neural network can be distracted from learning age traces. For this purpose, previous work is analyzed, re-implemented if required and experiments are conducted.