Barefoot Footprint Person Identification Algorithm Based on Feature Partitioning
In order to improve the accuracy of the barefoot footprint-based person identification algorithm,a deep learning-based foot recognition algorithm was proposed.The differing pressure experienced by different regions of the foot sole results in variations in the amount of information they contain.In order to obtain more stable and discriminative features,ResNet50 was utilized as the underlying network,and block processing was performed on the feature layer.A barefoot footprint database containing 2 000 individuals was constructed for training,and another database containing 3 000 individuals was used for testing.The algorithm achieves a top-1 recognition accuracy of 98.50％on the testing database using 1 000 test images from 500 individuals,surpassing the performance of the conventional ResNet50 network.It is observed in the experiment that the feature-based segmentation approach achieves excellent recognition results in barefoot footprint identification.
万方数据
维普
