GPU-Accelerated Feature Extraction for Real-Time Vision AI and LLM Systems Efficiency: Autonomous Image Segmentation, Unsupervised Clustering, and Smart Pattern Recognition for Scalable AI Processing with 6.6× Faster Performance, 2.5× Higher Accuracy, and UX-Centric UI Boosting Human-in-the-Loop Productivity
The DOI matches its Crossref record; 23 of the 23 references it cites were found in a registry or an index.
Abstract
The high computational cost of digital image processing, requiring high-performance hardware and extensive resources, severely limits real-time applications. While advancements in algorithm design and GPU acceleration have significantly improved efficiency, modern AI-driven applications such as large language models (LLMs), Generative AI (GenAI), medical imaging, autonomous vehicle perception, photography, advanced nano-scale semiconductor metrology, satellite image analysis, high-precision manufacturing, robotics, and real-time anomaly detection, still demand further optimization to reduce computational overhead and improve scalability.In this paper, we introduce GPU-Accelerated Feature Extraction to enhance runtime and efficiency in edge-based simulations. Our approach leverages AI-driven clustering, grouping images with similar visual and pattern characteristics to enable adaptive tuning on a small subset before generalizing across the full dataset. This method achieves a 3.78× reduction in runtime.Furthermore, rather than processing an entire image, we recognize and extract a single representative pattern or region of interest (ROI) per image, removing redundant data and background noise. This refinement results in an additional 1.74× runtime improvement, culminating in an overall 6.6× speed boost, enhancing Scalable Real-Time AI Processing. We also demonstrated that with a similar runtime, the accuracy achieved is 2.5× higher.This solution, integrated into Calibre SEMSuite™, supports multicloud and real-time deployment for enhanced scalability, usability, and performance, providing users with a powerful tool for fully automated, AI-driven image classification, making high-throughput image review feasible even at the scale required for cutting-edge applications.Beyond performance gains, this approach introduces autonomous data cleaning, anomaly detection and defect identification mechanism, allowing failed patterns and defective images to be identified without human intervention, boosting the reviewer productivity.As GenAI and LLM systems gain popularity, the computational demands on modern systems have reached unprecedented levels. As we demonstrate, thanks to feature extraction and ROI selection, instead of needing for the entire dataset to be processed, only a fraction of the data could be used. This is crucial for reducing the computational overhead of LLM systems.We demonstrate that our method enables high-precision, real-time AI inference with applications in computer vision, LLMs, autonomous systems, healthcare, and scalable AI computing.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“The study proves that careful UX design can lead to not only efficiency benefits but also to sustainable human–AI collaboration.”
From Human-in-the-Loop Artificial Intelligence: A Systematic Review of Concepts, Methods, and Applications (2026) · cites this paper as [110]
In the citing paper
“Ahi et al. developed a UX-centric human-in-the-loop system for product lifecycle management that achieved substantial improvements in reviewer productivity while maintaining decision quality [110].”
From Human-in-the-Loop Artificial Intelligence: A Systematic Review of Concepts, Methods, and Applications (2026) · cites this paper as [110]
Citations
Cites
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 1×
“Advancements in GPU acceleration and AI-driven feature extraction offer a promising solution to these challenges [2].”
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 2×
“By leveraging parallel computing capabilities, GPUs can significantly accelerate image analysis tasks, while AI-based clustering and feature selection techniques can optimize processing efficiency [1].”
Image quality assessment: from error visibility to structural similarity
Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. (2004) · IEEE Transactions on Image Processing · doi:10.1109/tip.2003.819861 · Cited by 57,892 · cited 2×
“Clustering images with similar illumination conditions ensures consistent processing ([5], [6]).”
A Computational Approach to Edge Detection
Canny, J. (1986) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.1986.4767851 · Cited by 30,004 · cited 1×
“To mitigate pixel-to-pixel noise inherent to imaging, a GPU-accelerated sliding window aggregates localized intensity values, enhancing edge extraction robustness ([7]).”
Multiresolution gray-scale and rotation invariant texture classification with local binary patterns
Ojala, T., Pietikainen, M., & Maenpaa, T. (2002) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2002.1017623 · Cited by 15,398
TensorFlow: A system for large-scale machine learning
Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., Devin, M., Ghemawat, S., Irving, G., Isard, M., et al. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1605.08695 · Cited by 11,822
An analysis of single-layer networks in unsupervised feature learning
Coates, A., Ng, A. Y., & Lee, H. (2011) · International Conference on Artificial Intelligence and Statistics · Cited by 4,675
A review of deep learning-based deformable medical image registration
Zou, J., Gao, B., Song, Y., & Qin, J. (2022) · Frontiers in Oncology · doi:10.3389/fonc.2022.1047215 · Cited by 126
Image Registration: Fundamentals and Recent Advances Based on Deep Learning
Chen, M., Tustison, N. J., Jena, R., & Gee, J. C. (2023) · Neuromethods · doi:10.1007/978-1-0716-3195-9_14 · Cited by 48
From IC Layout to Die Photograph: A CNN-Based Data-Driven Approach
Shao, H.-C., Peng, C.-Y., Wu, J.-R., Lin, C., Fang, S., Tsai, P.-Y., & Liu, Y.-H. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.3015469 · Cited by 45
Research works printed but not linked: 8
- González, 2018, Digital Image Processing
- Chen, 2017, Adaptive noise reduction for SEM images in sub-10nm lithography, IEEE Trans. Image Process., 26, 3876
- Ahi, 2022, Wafer image defect detection and characterization for manufacturing process calibration
- Yang, 2021, Contour-based defect detection for SEM images, IEEE J. Sel. Topics Signal Process., 15, 1134
- Lu, Deep learning for image registration
- Image Registration with Deep Learning — DeepReg Documentation
- Fenger, 2021, Wafer image denoising and contour extraction for manufacturing process calibration
- Ahi, 2025, Unsupervised, Scalable Clustering, Pattern Recognition, and Graphics Processing Unit (GPU)-Accelerated Contour Extraction from Challenging High-Variability Images Using Edge and High-Performance Computing (HPC) Architectures
Cited by
Lazaros, K., Vrahatis, A. G., & Kotsiantis, S. (2026) · Entropy · doi:10.3390/e28040377 · Cited by 50
Ahi, K., & Valizadeh, S. (2025) · Silicon Valley Cybersecurity Conference (SVCC) · doi:10.1109/svcc65277.2025.11133642 · Cited by 26
“• Efficient and Scalable AI Processing: Investigating architectures for unsupervised, scalable clustering and pattern recognition, potentially leveraging GPU-acceleration, edge, and HPC architectures for challenging high-variability image data relevant to security analytics [43], [67].”
Self-citationAhi, K., Hsieh, C.-H., & Fenger, G. L. (2025) · SPIE Photomask Technology Conference · doi:10.1117/12.3078485 · Cited by 2
“Our approach emphasizes GPU-accelerated learned feature representation , where deep neural networks are trained to automatically extract rich, compact, and semantically meaningful features directly from the raw input data [21].”
Self-citationHuang, B., Wu, Y., Wu, J., & Luo, L. (2026) · Chip · doi:10.1016/j.chip.2026.100197
Metadata
- Authors
- Ahi, K., Wu, S., Sriram, S., & Fenger, G.
- Affiliations
- Siemens (United States) (US)
- Siemens (Belgium) (BE)
- Siemens (Germany) (DE)
- Venue
- 2025 36th Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)
- Publisher
- IEEE
- Published
- 2025-05-05
- Pages
- 1-8
- ISSN
- 1078-8743 (ISSN-L)
- OpenAlex
- W4410779824
- Cited by
- 4
- References
- 23
- Citations by year
- 2026: 2 · 2025: 2
- FWCI
- 3.42
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- ASMC proceedings (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Computer Science and Engineering
- Keywords
- GPU acceleration, feature extraction, real-time image processing, image segmentation, unsupervised clustering, region of interest selection, edge computing, large language models, generative AI, defect detection, anomaly detection, image classification, runtime optimization
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: GPU-Accelerated Feature Extraction for Real-Time Vision AI and LLM Systems Efficiency: Autonomous Image Segmentation, Unsupervised Clustering, and Smart Pattern Recognition for Scalable AI Processing with 6.6× Faster Performance, 2.5× Higher Accuracy, and UX-Centric UI Boosting Human-in-the-Loop Productivity. https://app.cochat.ai/papers/10.1109/asmc64512.2025.11010527
- The paper
- Ahi, K., Wu, S., Sriram, S., & Fenger, G. (2025). GPU-Accelerated Feature Extraction for Real-Time Vision AI and LLM Systems Efficiency: Autonomous Image Segmentation, Unsupervised Clustering, and Smart Pattern Recognition for Scalable AI Processing with 6.6× Faster Performance, 2.5× Higher Accuracy, and UX-Centric UI Boosting Human-in-the-Loop Productivity. 2025 36th Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC). https://doi.org/10.1109/asmc64512.2025.11010527
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.