Automatic extraction of scale information for interactive measurement of anything in microscopy images

Shuo Meng; Shuai Zhang; Xinshuo Liang; Jinlian Hu

doi:10.1016/j.knosys.2025.113578

Automatic extraction of scale information for interactive measurement of anything in microscopy images

Shuo Meng, Shuai Zhang, Xinshuo Liang, Jinlian Hu^*

^*Corresponding author for this work

Department of Biomedical Engineering

Research output: Journal Publications and Reviews › RGC 21 - Publication in refereed journal › peer-review

Abstract

Quantitative analysis of microscopy images is crucial for objective measurement and a deep understanding of properties. Amid the surge in biomedical publications, the automatic extraction of scale information has become essential for achieving high-accuracy measurements, ensuring reproducible research, and meeting the high-throughput demands of data mining. However, this vital aspect is often neglected in existing studies. To address this gap, we present an automated system combining two key innovations: (1) YOLO-OCR, a unified trainable model for simultaneous object detection, text detection, and text recognition, and (2) the largest dataset of annotated scale bars with 14,000 real and 5,726 synthesized scientific images. Experiments show that our solution achieves higher precision for scale bar detection (98.0% vs. 96.4% baseline) and scale label recognition (96.6% vs. 93.6% baseline), and reduces computational costs by 5% (44.2 ms vs. 46.5 ms baseline). Furthermore, an online system integrated with Segment Anything for Microscopy (MicroSAM) is developed to help researchers conveniently measure and quantitatively analyze microscopy images. © 2025

Original language	English
Article number	113578
Journal	Knowledge-Based Systems
Volume	324
Online published	3 Jun 2025
DOIs	https://doi.org/10.1016/j.knosys.2025.113578
Publication status	Online published - 3 Jun 2025

Funding

The authors acknowledge the financial support from the National Natural Science Foundation of China (NSFC) titled \u201CStudy of High Performance Fiber to be Achieved by Mimicking the Hierarchical Structure of Spider-Silk\u201D; grant no. 52073241, \u201C Study of Multi-Responsive Shape Memory Polyurethane Nanocomposites Inspired by Natural Fibers, Hong Kong Special Administrative Region of China \u201D; grant no. 51673162 , \u201C Developing Spider-Silk-Model Artificial Fibers by a Chemical Synthetic Approach, Hong Kong Special Administrative Region of China \u201D; grant no. 15201719, the Collaborative Research Fund with the title of \u201CFundamental Study towards Real Spider Dragline Silk Performance through Artificial Innovative Approach\u201D; project no. 8730080, the startup grant of CityU titled \u201C Laboratory of Wearable Materials for Healthcare, Hong Kong Special Administrative Region of China \u201D; grant no. 9380116, and the contract research titled \u201CDevelopment of breathable fabrics with nano-electrospun membrane\u201D; CityU ref.: 9231419.

Research Keywords

Interactive measurement
Object detection
Scale bar detection
Scale information extraction
Text recognition

Access Output

10.1016/j.knosys.2025.113578

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

@article{a18aa5ecc61b4bb4abcc5c4e3273f455,

title = "Automatic extraction of scale information for interactive measurement of anything in microscopy images",

abstract = "Quantitative analysis of microscopy images is crucial for objective measurement and a deep understanding of properties. Amid the surge in biomedical publications, the automatic extraction of scale information has become essential for achieving high-accuracy measurements, ensuring reproducible research, and meeting the high-throughput demands of data mining. However, this vital aspect is often neglected in existing studies. To address this gap, we present an automated system combining two key innovations: (1) YOLO-OCR, a unified trainable model for simultaneous object detection, text detection, and text recognition, and (2) the largest dataset of annotated scale bars with 14,000 real and 5,726 synthesized scientific images. Experiments show that our solution achieves higher precision for scale bar detection (98.0% vs. 96.4% baseline) and scale label recognition (96.6% vs. 93.6% baseline), and reduces computational costs by 5% (44.2 ms vs. 46.5 ms baseline). Furthermore, an online system integrated with Segment Anything for Microscopy (MicroSAM) is developed to help researchers conveniently measure and quantitatively analyze microscopy images. {\textcopyright} 2025",

keywords = "Interactive measurement, Object detection, Scale bar detection, Scale information extraction, Text recognition",

author = "Shuo Meng and Shuai Zhang and Xinshuo Liang and Jinlian Hu",

year = "2025",

month = jun,

day = "3",

doi = "10.1016/j.knosys.2025.113578",

language = "English",

volume = "324",

journal = "Knowledge-Based Systems",

issn = "0950-7051",

publisher = "Elsevier B.V.",

}

TY - JOUR

T1 - Automatic extraction of scale information for interactive measurement of anything in microscopy images

AU - Meng, Shuo

AU - Zhang, Shuai

AU - Liang, Xinshuo

AU - Hu, Jinlian

PY - 2025/6/3

Y1 - 2025/6/3

N2 - Quantitative analysis of microscopy images is crucial for objective measurement and a deep understanding of properties. Amid the surge in biomedical publications, the automatic extraction of scale information has become essential for achieving high-accuracy measurements, ensuring reproducible research, and meeting the high-throughput demands of data mining. However, this vital aspect is often neglected in existing studies. To address this gap, we present an automated system combining two key innovations: (1) YOLO-OCR, a unified trainable model for simultaneous object detection, text detection, and text recognition, and (2) the largest dataset of annotated scale bars with 14,000 real and 5,726 synthesized scientific images. Experiments show that our solution achieves higher precision for scale bar detection (98.0% vs. 96.4% baseline) and scale label recognition (96.6% vs. 93.6% baseline), and reduces computational costs by 5% (44.2 ms vs. 46.5 ms baseline). Furthermore, an online system integrated with Segment Anything for Microscopy (MicroSAM) is developed to help researchers conveniently measure and quantitatively analyze microscopy images. © 2025

AB - Quantitative analysis of microscopy images is crucial for objective measurement and a deep understanding of properties. Amid the surge in biomedical publications, the automatic extraction of scale information has become essential for achieving high-accuracy measurements, ensuring reproducible research, and meeting the high-throughput demands of data mining. However, this vital aspect is often neglected in existing studies. To address this gap, we present an automated system combining two key innovations: (1) YOLO-OCR, a unified trainable model for simultaneous object detection, text detection, and text recognition, and (2) the largest dataset of annotated scale bars with 14,000 real and 5,726 synthesized scientific images. Experiments show that our solution achieves higher precision for scale bar detection (98.0% vs. 96.4% baseline) and scale label recognition (96.6% vs. 93.6% baseline), and reduces computational costs by 5% (44.2 ms vs. 46.5 ms baseline). Furthermore, an online system integrated with Segment Anything for Microscopy (MicroSAM) is developed to help researchers conveniently measure and quantitatively analyze microscopy images. © 2025

KW - Interactive measurement

KW - Object detection

KW - Scale bar detection

KW - Scale information extraction

KW - Text recognition

UR - http://www.scopus.com/inward/record.url?scp=105007413344&partnerID=8YFLogxK

UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-105007413344&origin=recordpage

U2 - 10.1016/j.knosys.2025.113578

DO - 10.1016/j.knosys.2025.113578

M3 - RGC 21 - Publication in refereed journal

SN - 0950-7051

VL - 324

JO - Knowledge-Based Systems

JF - Knowledge-Based Systems

M1 - 113578

ER -

Automatic extraction of scale information for interactive measurement of anything in microscopy images

Abstract

Funding

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

GRF: Developing Spider-Silk-Model Artificial Fibers by A Chemical Synthetic Approach

Cite this

Automatic extraction of scale information for interactive measurement of anything in microscopy images

Abstract

Funding

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Projects

GRF: Developing Spider-Silk-Model Artificial Fibers by A Chemical Synthetic Approach

Cite this