2P1-B06 Image Processing of Particle Detection for Asbestos Qualitative Analysis Support Method : Particle Detection Based on Color Variance of Background Area

Kenichi Ishizu, Hiroshi Takemura, Kuniaki Kawabata, Hajime Asama, Taketoshi Mishima, Hiroshi MizoguchiThe Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec) 2008 · DOI

Abstract

We propose a technique for automatically classifying background areas in microscopic images by calculating color variance in RGB space. This approach enables background classification that is robust to changes in image brightness and color. The validity of the technique was demonstrated through particle extraction and counting on images taken with a phase-contrast microscope. Our goal is to develop an automated particle detector for the "dispersion staining method" used in asbestos analysis.

Introduction

Asbestos qualitative analysis is a critical but labor-intensive task. In the "dispersion staining method," human inspectors must manually count approximately 3,000 particles per sample using a phase-contrast microscope. This process is not only time-consuming but also physically and mentally exhausting.

While quantitative analysis has seen some automation, qualitative analysis remains a bottleneck. The primary challenge for computer vision in this field is the inconsistency of images: changes in microscope settings, lighting, and sample preparation mean that the background color and brightness shift constantly. Standard techniques like simple background subtraction often fail because they cannot adapt to these fluctuations.

Methodology

The proposed system employs a two-step process: Background Identification followed by Particle Extraction.

RGB Variance Classification

Instead of relying on absolute pixel values, the system divides a 630x480 pixel image into small 30x30 blocks. For each block, it plots the distribution of all pixels in RGB space. The core insight is that in areas containing particles, the RGB values are widely scattered (high variance), whereas in pure background areas, the RGB values cluster tightly together (low variance).

RGB distribution of a background area showing tight clustering, which indicates low variance.

Dynamic Background Modeling

By identifying these low-variance blocks, the system calculates an average RGB value for the background of that specific image. This allows the algorithm to be "self-calibrating" for every new image it processes, handling local variations in brightness across the field of view. The system creates a model background based on these averages and subtracts it from the original image to isolate particles for counting.

Results and Discussion

The study used a dataset of 15 microscopic images captured using a Nikon ECLIPSE 80i phase-contrast microscope. The results were benchmarked against manual counts performed by a professional inspector.

Accuracy and Stability

The proposed method achieved a 101.7% correlation with the human inspector (549 particles detected vs. 540 by the inspector). More importantly, the method demonstrated significant stability compared to traditional background subtraction.

Comparison of particle counting rates between the proposed technique and standard background subtraction across 15 test images.

As shown in the performance data, the proposed technique stayed within a ±10% error margin for 12 out of 15 images. In contrast, the standard background subtraction method showed massive fluctuations, with accuracy swinging wildly depending on the brightness of the specific frame. This proves that background classification via variance is the critical step for reliable automated counting.

Conclusion

This research marks a significant step toward relieving the burden on lab technicians by demonstrating that statistical color analysis is a viable path for automating hazardous material inspection. Future work will involve testing on a wider variety of asbestos types and concentrations, as well as optimizing the system for real-time processing and handling overlapping particles.

Summary updated

Abstract

We propose the technique of classifying the background area automatically by calculating the color variance in the RGB space of the picture. By using this technique, the background area classification robust to change of the brightness and color of the picture was enabled. Moreover, the validity of this technique was shown by actually performing particle extraction and verifying the result which carried out calculation. We aim at development of the particle detector which performs particle counting automatically about the "distributed staining method" which is one of the processing of asbestos analysis. The pictures were taken by the phase microscope.
BibTeX
@inproceedings{ishizu2008p,
  author = {Kenichi Ishizu and Hiroshi Takemura and Kuniaki Kawabata and Hajime Asama and Taketoshi Mishima and Hiroshi Mizoguchi},
  title = {{2P1-B06 Image Processing of Particle Detection for Asbestos Qualitative Analysis Support Method : Particle Detection Based on Color Variance of Background Area}},
  booktitle = {The Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec)},
  year = {2008},
  doi = {10.1299/jsmermd.2008.\_2p1-b06\_1},
}
Figure