Cross-Domain Footprint Image Retrieval Based on Feature Chunking and Domain Fusion

Existing image retrieval methods face challenges in extracting highly discriminative domainshared features from cross-domain footprint images. Therefore, this paper proposes a cross-domain footprint image retrieval method based on feature chunking and inter-domain fusion. Firstly, global features of the footprint images are extracted using ResNet50 as the backbone network. Secondly, a horizontal block feature extraction approach is employed to obtain more discriminative features. Finally, a cross-domain feature fusion method is utilized to extract domain-shared information, and an equilibrium loss is designed to optimize the fusion features. The proposed method is evaluated on the self-collected dataset of 200 human cross-domain footprint image. Result indicate that Rank-1 accuracy achieves 91.38% and 84.50% for optics retrieval pressure and pressure retrieval optics, respectively.