Autofocus methods based on laser illumination.

Area of Science:

• Optics and Photonics
• Microscopy
• Machine Learning for Imaging

Background:

• Autofocusing is critical for microscopic measurement, but natural-image methods struggle with accuracy and robustness due to object diversity.
• Existing autofocusing techniques face limitations in precision and reliability when dealing with varied microscopic samples.

Purpose of the Study:

• To introduce a high-precision autofocus method using laser illumination, termed laser split-image autofocus (LSA).
• To develop and evaluate a lightweight comparative framework model (SCM) for LSA to enhance focusing accuracy and robustness.
• To establish a benchmark for LSA methods through quantitative analysis and a custom dataset.

Main Methods:

• Proposed the laser split-image autofocus (LSA) method, which actively introduces image features using laser illumination.
• Quantitatively analyzed and evaluated common non-learning-based and learning-based methods for LSA.
• Developed a lightweight comparative framework model (SCM) for LSA and created a realistic split-image dataset for training.

Main Results:

• LSA demonstrated superior focusing performance compared to natural-image-based autofocus methods.
• The proposed SCM significantly improved accuracy and robustness over existing learning and non-learning methods.
• SCM achieved a mean focusing error of 0.317µm in complex industrial measurement scenes.

Conclusions:

• LSA offers a more effective approach to autofocusing in microscopic measurements.
• The SCM provides a substantial advancement in precision and robustness for laser-based autofocusing.
• SCM is highly suitable for demanding industrial measurement applications requiring high accuracy.