Research On Facial Expression Generation And Recognition Based On Deep Learning

Facial expression generation and recognition is to make the input face generate a specific facial expression and identify which category the expression belongs to through image processing algorithms.It is a hot spot in the current face analysis research.Virtual reality,animation character production and other fields have broad application scenarios and important value.Although great progress has been made,current facial expression generation and recognition algorithms still have limitations.For example,facial expression generation algorithms produce a large number of artifacts and blurs when dealing with large-angle and large-gap facial expression generation tasks.The facial expression recognition algorithm needs a lot of additional data to improve the generalization performance of the model,but the annotated expression images are difficult to obtain.And the lightweight face detection algorithm is not good for the detection of difficult cases.The main work of this paper is summarized as follows:(1)First,we improve the first step of face expression generation and recognition task-face detection,and innovatively propose a lightweight face detection algorithm,which trains samples from easy to hard and add an additional anchor-free branches.The experimental results show that the algorithm proposed in this paper can be improved compared with the baseline methods on both WIDER FACE and FDDB datasets.In particular,the lightweight network achieved 87.9%AP on the WIDER FACE hard set with only 1.3 5G FLOPs.(2)A novel landmarks guided attentive generative adversarial network is proposed,which includes a landmarks-guided attentive matrix and a matching generative adversarial network.Qualitative and quantitative experiments can prove that the algorithm proposed in this paper can achieve a certain improvement in large-angle and large-gap face generation task,which shows the superiority of its network structure and novel attention mechanism.(3)A generative adversarial network-assisted facial expression recognition is proposed.With the help of the above-mentioned novel generative adversarial network,the method proposed in this paper can expand the training set of facial expression recognition without introducing other datasets,thereby improving the model’s performance.The experimental results show that the algorithm can improve the performance of the facial expression baseline.