Exploring the Impact of Color Space in Malaria Parasite Detection using Deep Learning

Malaria, a life-threatening infection caused by Plasmodium, remains a major concern in tropical regions. Using the MP-IDB thin-smear image set, this study examines how color transformations influence the detection of malaria parasites at different developmental stages. Two state-of-the-art detectors—YOLOv8 and Faster R-CNN—were trained with several preprocessing variants: LAB, CMY, and HSL color spaces, plus feature-engineering techniques such as Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP). Performance was assessed with mAP50, mAP50-95, precision, and recall, with particular attention to small, artifact-obscured parasites that have challenged earlier models.Results reveal that the LAB color space yields the most substantial gains. For YOLOv8, LAB lifted mAP50 from 0.612 to 0.687; for Faster R-CNN, it raised mAP50 from 0.482 to 0.546. LAB’s advantage likely stems from its superior separation of luminance and chromaticity, enabling networks to isolate informative features more effectively. In contrast, traditional feature-extraction add-ons (HOG, LBP, ORB) provided negligible benefit, suggesting that deep networks’ internal filters already capture the relevant spatial patterns.These findings highlight color-space selection as a simple yet powerful lever for boosting malaria parasite detection, especially for tiny targets. Future work should investigate spectrum-based augmentations, color normalization, channel expansion, and bespoke architectures to further improve robustness across lighting variations and morphological diversity.