Hybrid contour and geometric partitioning for accurate plantar foot region segmentation.

BACKGROUND

Precise segmentation of plantar foot regions is crucial for analyzing foot structure and pressure distribution, aiding in the diagnosis of pathologies and enabling preventive interventions. However, conventional segmentation approaches often struggle to accurately delineate key anatomical regions and detect their boundaries, particularly in the presence of foot abnormalities.

METHODS

We created a dataset of plantar pressure images and proposed a hybrid algorithm that integrates edge contour detection techniques with dynamic geometric partitioning to address persistent challenges in plantar region segmentation. Our method first determines the lengths of the left and right feet using precise contour detection, then partitions the plantar surface into primary anatomical regions (forefoot, midfoot, and heel) based on standardized geometric proportions. Additionally, the methodology allows for finer subdivisions (e.g., inner/outer forefoot) that adapt to the unique morphology of each foot. This algorithm accommodates five foot types, including normal, low arch, high arch, inward heel tilt, and outward heel tilt.

RESULTS

A comparative evaluation of three edge detection methods revealed that the Canny algorithm, when combined with geometric partitioning, yielded superior performance. On a dataset of 200 plantar pressure footprints encompassing both normal and abnormal feet, this hybrid approach achieved Intersection over Union (IoU) and mean Average Precision (mAP) scores exceeding 0.90 across all segmented regions (forefoot, midfoot, and heel). Furthermore, the results indicate that the proposed hybrid algorithm performs comparably across both normal and abnormal foot types, with no significant differences observed.

CONCLUSIONS

Our synergistic integration of contour detection and geometric partitioning yields an efficient technique for segmenting plantar regions from static plantar pressure images. Validation on a diverse dataset shows that the proposed approach accurately distinguishes foot-specific regions across five different foot types, including both normal and pathological cases.