Tongue diagnosis serves as a cornerstone of Traditional Chinese Medicine (TCM), where pathological changes in internal organs can be inferred through systematic analysis of regional tongue features. Recent advances in computer vision have enabled quantitative health assessment through automated tongue image analysis. However, critical challenges persist in clinical implementation, particularly regarding inconsistent imaging conditions (e.g., illumination direction, intensity variations, and device heterogeneity), which significantly compromise diagnostic reliability. This study presents a comprehensive computational framework for robust tongue image analysis, addressing two fundamental aspects: (1) imaging standardization and (2) clinically interpretable feature extraction. Our technical contributions are threefold: First, we propose a Standard Deviation-Weighted Gray World (SDWGW) algorithm for illumination normalization, effectively reducing inter-sample variability across diverse acquisition conditions. Second, we develop a hierarchical segmentation pipeline combining an FCN-16s deep model for precise tongue region extraction with anatomically grounded zoning (tip, middle, and edges) aligned with TCM diagnostic principles. Third, we establish a multi-label classification system using EfficientNet-B0 to quantify five diagnostically critical features: tongue body color, coating color, coating thickness, tooth marks, and fissures. Extensive experiments were conducted to evaluate our proposed method, and the results demonstrate its effectiveness in multi-label classification tasks and challenging conditions.