Contextual and structural representation learning for fine-grained barefoot footprint region recognition

Barefoot footprints provide valuable biometric evidence for forensic investigation and suspect identification. However, the rich structural details and diverse scale distributions of footprint images pose significant challenges for accurate fine-grained region recognition. Existing methods often struggle to effectively model contextual information and structural characteristics, limiting their ability to characterize subtle morphological variations. To address these issues, this paper proposes a robust framework for contextual-structural representation learning in barefoot footprint region recognition. A context-guided spatial feature reconstruction pyramid network (CGSFRPN) is introduced to enhance semantic consistency and feature interactions across different scales. To improve directional perception and edge detail modeling, a Direction-Aware Pinwheel Convolution (DAPConv) is incorporated to capture structural patterns with diverse orientations. Furthermore, a ConvFormer module based on the MetaFormer architecture is employed to facilitate efficient semantic representation and cross-region information integration. By jointly exploiting contextual information and structural characteristics, the proposed framework achieves robust recognition of fine-grained footprint regions, including toes, arches, and heels. Experiments conducted on a proprietary barefoot footprint dataset demonstrate that the proposed method achieves a mean Average Precision (mAP) of 83.2%, outperforming several mainstream object detection approaches and exhibiting superior visualization performance. These results highlight the effectiveness and robustness of the proposed framework for fine-grained barefoot footprint analysis and its potential applications in intelligent forensic identification.