Occluded Face Recognition Algorithm Incorporating Attention Mechanism

Face recognition technology is widely used in various industries with its high recognition rate and the market size is growing. However, the problem of face recognition with occlusion has not been completely solved, and how to solve the problem of low accuracy of face recognition under real scene occlusion is still one of the hotspots in today's research. FaceNet face recognition algorithm is widely used in industry because of its fast operation speed, and its network structure can be flexibly improved for different situations. Therefore, this paper proposes an occluded face recognition algorithm incorporating the attention mechanism, which improves the network structure by combining the attention mechanism on the basis of the original FaceNet network. Firstly, the original backbone network is replaced with a MobileNetV3 network with its own Squeeze-and-Excitation channel attention mechanism, so that the network focuses on the unobstructed face region. Then, the attention mechanism is further improved. The original SE attention mechanism is changed to a coordinate attention (CA) mechanism that can embed location information in the channel attention to further enhance the more important feature information. The experimental results show that the improved algorithm achieves good results, with an accuracy of 99.61% on the LFW face dataset, and the accuracy on the LFW_Mask masked face dataset is also significantly improved, which can achieve accurate recognition of masked faces.