Cross-modality Deformed Footprint Image Retrieval by Shape and Multi-layer Information

Deformed of the surface can increase the modality discrepancy between barefoot footprint dust images and grayscale images. To address this issue, this paper proposes a crossmodality footprint image retrieval algorithm based on shape and multi-layer information. Firstly, a cross-modality dataset containing barefoot footprint images from 100 individuals is collected and constructed. Shape features are extracted from the footprint grayscale images to reduce the interference of deformation features, thereby minimizing the modality discrepancy between the two types of images. Then, hierarchical fusion (HF) module is further proposed to capture important information across different layers and to focus on the relationships among the toe, sole, and heel regions of the footprint, reduce the discrepancy between the two modalities under the combined constraints of ID loss, triplet loss, and center guided separation loss. Extensive experiments conducted on the footprint dataset constructed for this study demonstrate that the retrieval accuracy of the proposed method surpasses that of several other retrieval methods, showing significant promise for the comparison and identification of deformed surface footprints.