Lip pattern recognition algorithm based on attention mechanism and multi-scale features

Abstract

For the preliminary exploration of the deep neural network in the lip pattern recognition, a CNN-based lip pattern recognition algorithm is proposed. The lightweight network MobileNetV2 is more suitable for lip print recognition, but the recognition accuracy of this model needs to be improved compared with the existing machine learning recognition algorithms. The attention mechanism module and the grouped multi-scale feature fusion module are designed to not only increase the model’s ability to finely represent texture features, extract features, and adapt to different input scales but also to further compress the number of parameters in the network model. The ablation experimental results show that the improved recognition model achieves 98.56% recognition accuracy. The effectiveness of the improved algorithm is finally verified by the experiments.