Selecting a hijab color that harmonizes with facial skin tone is critical for enhancing visual aesthetics and individual confidence. However, color selection in retail stores is conventionally conducted through subjective visual guessing and repetitive physical trials, which leads to operational inefficiencies and color mismatch. This study presents an embedded digital image processing framework deployed on a single-board computer (Raspberry Pi 4) to automate facial skin tone acquisition, undertone classification, and hijab color recommendation in real time. The methodology establishes an end-to-end pipeline comprising image acquisition with a standardized USB webcam, multi-stage Region of Interest (ROI) extraction using automated facial landmarking and forehead/cheek segmentation, illumination compensation via Gray-World white balancing, and quantitative color space transformation. A comparative analysis across RGB, HSV, and CIE L*a*b* spaces was executed, establishing an empirical decision-boundary threshold model based on individual channel distributions (a* and b* chromatic components and R/(G+B) ratios) to categorize skin undertones into Warm, Cool, and Neutral. Testing on 120 diverse demographic datasets across varied illumination conditions (250–650 lux) demonstrated an undertone classification accuracy of 88.33% with an F1-score of 0.88, while user acceptance and expert stylist evaluation yielded an 89.2% satisfaction rate. Running on an embedded Flask architecture, the system maintains a processing latency of 142 ms (18.4 FPS) and low resource utilization. This research provides an objective, reproducible, and computationally efficient edge-computing solution for real-time personalized cosmetic and fashion recommendations.