Performance evaluation of low-resolution lip print recognition algorithm
In order to explore and research on network models for lip print recognition criminal investigations,eight different CNN models are selected and introduced from the aspects of network structure design,core modules and the connection between networks,and the performance of different network models is evaluated on the created low-resolution lip print database.At the same time,experiments are also carried out with different learning rates and network layers.The experimental results show that the lightweight model MobileNetV2 achieves a recognition rate of 97.22%,and its recognition effect is the best,and model size is only 8.63 MB.It is verified through experiments that the recognition algorithm based on the CNN models can also be well applied to the lip print recognition task,which effectively makes up for the shortcomings of the traditional recognition algorithm.
万方数据
维普
