Research and Improvement of Image Sharpness Evaluation and Autofocus Method

Introduction: Accurate focus evaluation is vital in optical imaging systems for microstructure fabrication and defect inspection. However, existing sharpness metrics often show low sensitivity, multiple local extrema, and poor noise robustness, impairing autofocus performance. Methods: We propose TGA, a novel focus measure combining local variance with gradient-based evaluation. Images are first preprocessed using anisotropic diffusion filtering to smooth homogeneous areas while preserving edges. Then, the Tenengrad method, based on Sobel gradients, is combined with a local gray-level variance operator that measures the deviation of each pixel from its neighborhood mean. This enhances detail detection and suppresses noise. Finally, results are normalized to reduce background and spurious interference. Results: Experiments on diverse datasets show that TGA significantly outperforms traditional focus metrics. Its peak sharpness response is ~3.7 times higher, with steeper focus curves and improved signal-to-noise separation under both additive and speckle noise. Quantitative metrics confirm TGA's superior performance. Discussion: TGA’s improved sensitivity and robustness enable more accurate focus determination, minimizing false peaks due to noise. Though anisotropic diffusion and variance calculations add computational load, parameters can be optimized offline, and real-time implementation is feasible. Some parameter tuning is required for specific imaging conditions. Conclusion: The TGA method enhances sharpness evaluation by nearly 3.7 times, offering a strong peak response and robust noise resistance, which provides practical benefits for automated focusing in precision imaging tasks, such as microfabrication and defect inspection.