With the development of deep learning methods, the accuracy and speed of face recognition based on visible light in ideal environments have reached an excellent level. However, in complex environments such as low light, the lack of a light source keeps visible images from reflecting face details, resulting in reduced or even invalid face recognition. Aiming at the problems in this issue, a face recognition method in complex environments based on infrared-visible light fusion is proposed. Firstly, an infrared and visible fusion recognition network combining CNN and Transformer is introduced, specifically designed for low illumination environments. This network combines CNN and visual Transformer in parallel to form a single-mode feature fusion module, which is utilized to effectively utilize local details and global context information from the source image. Additionally, a multimodal feature fusion strategy based on the average difference of modes is proposed to enhance the distinctive expression of different regional features in the source image. Secondly, a lightweight face recognition network MobileFaceNet-Coo and an adaptive recognition strategy based on edge-cloud collaboration are proposed in order to solve the problem of large and slow fusion recognition network models in practical applications. This strategy selects the recognition model through image quality and effectively utilizes hardware resources. Experimental results demonstrate that the recognition rate of fused infrared light is 13.96 percentage point higher than that of visible light alone. Real-world project result shows that this method significantly improves real-time and accuracy of face recognition in complex environments.

Portrait semantic segmentation is one of the important research contents in the field of computer vision, but the existing portrait semantic segmentation methods are liable to ignore the small size portraits in multi-person portrait images. At the same time, the segmentation results are prone to the phenomenon of mutual adhesion between multiple portraits. Moreover, the phenomenon of mutual occlusion between portraits in the image easily leads to poor segmentation accuracy of portrait edges. Based on the above problems, a semantic segmentation method for multiple portraits with fused label semantics is propose, where multiple labels are assigned to multiple portraits in an image, and semantic labels are embedded as inputs to the encoder at the same time, and the semantic labels and the image feature representations are correlated using the cross-modal cross-attention module, and the semantically fused feature representations are obtained as outputs of the encoder at each layer of the model. The HRF attention module is proposed to generate multiple hypotheses for image based on target detection algorithm for feature extraction separately. The network is trained and tested on Supervisely augmented dataset. The experimental results show that the algorithmic model achieves 95.94%, 94.60%, and 96.02% accuracy on the three evaluation metrics of PA, MIoU, and Dice, respectively, and has higher segmentation accuracy than the semantic segmentation models U-net, PSPNet, Deeplab v3+, PortraitNet, and Swin Unet.