Training-Free White Blood Cell Detection: Achieving State-of-the-Art Performance with Traditional Computer Vision

초록

This study presents a training-free approach for white blood cell (WBC) detection using traditional computer vision on the BCCD dataset. While deep learning methods dominate blood cell analysis, they require extensive training data and computational resources, limiting clinical adoption. Our methodology combines HSV color space preprocessing, watershed segmentation, and morphological operations to achieve 99.7% mAP@0.5 with sub-6ms processing times. The approach eliminates training requirements while providing interpretable results suitable for clinical environments. Results demonstrate competitive performance compared to AI-based methods with significant advantages in deployment simplicity and computational efficiency. © 2025 IEEE.

키워드