基于特征分块的赤足足迹人身识别算法

Barefoot Footprint Person Identification Algorithm Based on Feature Partitioning

为了提高赤足足迹人身识别算法的准确率,提出了一种基于深度学习的足迹识别算法。足底各区域所受压力的不同导致了它们包含的信息量存在一定的差异性,为了获取更稳定、区分度更高的特征,该算法采用ResNet50作为基础网络,在特征层进行分块处理。构建了一个包含2000人的赤足足迹库进行训练和一个包含3000人的赤足足迹库进行测试,该算法利用500人1000幅测试图在测试库上首位识别准确率达到了98.50%,优于常规的ResNet50网络。实验表明,基于特征分块的足迹识别算法在赤足足迹识别中获得了很好的识别效果。

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 2000 individuals was constructed for training,and another database containing 3000 individuals was used for testing.The algorithm achieves a top-1 recognition accuracy of 98.50%on the testing database using 1000 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.