High Recognition Ratio Image Processing Algorithm of Micro Electrical Components in Optical Microscope

Jie Wu, Zuren Feng, Lei WangTELKOMNIKA (Telecommunication Computing Electronics and Control) 2014 · DOI

TL;DR

This paper presents a robust image processing pipeline for automatically identifying and qualifying micro-electrical components, specifically chip resistors, captured under a microscope. By combining median filtering for noise reduction, lateral inhibition for sharpening, adaptive thresholding for segmentation, and a novel area-based qualification criterion, the method achieves high recognition ratios and reliable quality testing for these tiny, often challenging-to-image components.

What Problem Does This Paper Address?

The increasing miniaturization of electronic components, driven by the trend of integration, presents a significant challenge for quality control. These micro-electrical components are too small to be inspected manually with sufficient accuracy. While optical microscopes can provide magnification, the captured images often suffer from low contrast, blurriness, and noise due to factors like uneven lighting and diffraction. This makes it difficult to reliably extract and assess the quality of individual components. The paper aims to address this by developing an automated image processing algorithm that can accurately identify, segment, and qualify these tiny components from microscopic images.

Methods

The proposed image processing pipeline consists of several key stages:

  • Preprocessing and Target Enhancement:
    • Median Filtering: This non-linear filtering technique is employed to suppress noise while preserving image details, which is crucial for small components that can easily be obscured by noise.
  • Lateral Inhibition Sharpening: Inspired by human visual systems, this algorithm sharpens target edges by enhancing differences between neighboring pixels. This helps to delineate the component from the background and improve contour visibility.
  • Target Region Extraction:
    • Adaptive Threshold Segmentation: Instead of a single global threshold, this method uses local thresholds calculated for sub-images. This approach is more robust to variations in lighting and contrast across the image, leading to more reliable extraction of target regions.
  • Target Distinction and Qualification:
    • Mark Clustering Algorithm: This algorithm is used to distinguish and label multiple targets within the image based on their characteristics.
  • Area-Based Qualification: For chip resistors, the paper proposes a novel qualification method. Instead of relying on a single measurement, it compares the pixel area values of the solder regions on both sides of the chip resistor. If these areas are within a ±10% tolerance of each other, the component is deemed qualified. This accounts for potential inaccuracies in image capture and positioning.

Datasets

The paper does not explicitly mention the use of pre-existing benchmark datasets. Instead, it utilizes images of chip resistors captured using a microscope in a production process setting. Figure 5 shows an example of an "Original image under microscope," and subsequent figures illustrate the results of the processing pipeline on these captured images.

Key Insights and Ideas

  • Median Filtering's Advantage: The paper highlights that median filtering is superior to simple smoothing filtering for this application because it effectively removes noise without excessively blurring image details, which is critical for identifying small components.
  • Lateral Inhibition for Edge Enhancement: The use of lateral inhibition is shown to be more effective than conventional Sobel operators for sharpening edges and distinguishing adjacent targets, leading to clearer contours.
  • Adaptive Thresholding for Robust Segmentation: The adaptive threshold segmentation algorithm proves crucial for handling the non-uniform lighting and contrast variations often encountered in microscopic images, ensuring reliable extraction of target regions.
  • Area Ratio for Qualification: The proposed method of comparing the solder region areas (A1 and A2) with a tolerance is a practical and effective way to qualify chip resistors, accounting for real-world capture variations and improving the accuracy of the quality test.

Key Results

The experimental results demonstrate that the proposed image processing algorithm and qualified detection algorithm are effective. The paper shows that:

  • The image quality is significantly improved after preprocessing steps like median filtering and lateral inhibition sharpening.
  • Adaptive threshold segmentation successfully segments target regions from the background.
  • The mark clustering algorithm effectively distinguishes multiple targets.
  • The area-based qualification method provides a reliable means to assess chip resistor quality.
  • The overall algorithm effectively improves the accurate recognition ratio, providing a reliable method for qualified testing. While specific quantitative metrics like accuracy percentages are not detailed, the qualitative results and the authors' claims suggest a high level of performance.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that the captured images from the microscope can exhibit low contrast and poor clarity, which the proposed algorithms aim to mitigate.
  • The magnification and capture angle of the microscope are noted as being uncertain, which is a factor addressed by the area-based qualification method.
  • The paper does not provide extensive quantitative performance metrics (e.g., precision, recall, F1-score) for the entire pipeline.
  • Future Work:
    • The authors suggest that the developed algorithm can be applied to image processing and other similar problems, implying potential for broader applications beyond chip resistor inspection.
  • The paper's focus is on a specific component (chip resistor), and future work could explore its applicability to a wider range of micro-electrical components with varying characteristics.

Summary updated

Abstract

In order to process small components of original image under the microscope, firstly, this paper adopts median filtering algorithm to enhance targets; and the targets are sharpened by using lateral inhibition algorithm, the edge of targets is outlined. In order to get reliable target region, adaptive threshold segmentation algorithm is used to extract need target region, and characteristics of target is used to distinguish multiple targets. Based on the chip resistor, one tiny component, in the captured image, we judge if the chip resistor is qualified by calculating the pixels area values. The experimental results show that, the image processing algorithm and qualified detection algorithm is reasonable, which provides the theoretical basis and implementation method of effective target extraction and further qualified test.
BibTeX
@inproceedings{wu2014high,
  author = {Jie Wu and Zuren Feng and Lei Wang},
  title = {{High Recognition Ratio Image Processing Algorithm of Micro Electrical Components in Optical Microscope}},
  booktitle = {TELKOMNIKA (Telecommunication Computing Electronics and Control)},
  year = {2014},
  doi = {10.12928/telkomnika.v12i4.304},
}
Figure