Research On Semantic Information Extraction And Coding Algorithms Of Face Expression

In the face of the challenges posed by the exponential growth of global traffic and the limitations of conventional communication technologies,there is a growing interest in task-oriented semantic communication.Deep learning has emerged as a key enabler for semantic communication,and its combination with semantic communication is a highly debated topic in the field of communication.A facial expression-based semantic information extraction and coding algorithm can significantly enhance the accuracy of facial expression recognition and enable intelligent products to better comprehend human emotions during communication.This paper presents innovative research conducted around the semantic information extraction and coding algorithm for facial expressions in the context of semantic communication.(1)In the area of affective computing and human-computer interaction,there is a strong relationship between the tasks of recognizing facial expressions and identifying facial action units.However,both tasks often neglect the shared semantic information present in facial expressions.This study proposes a transfer learning model that considers facial expression recognition as the source task,and facial action unit recognition as the target task.This is driven by real factors such as the availability of more annotated data for facial expression recognition and the easier trainability of facial expression recognition models compared to facial action unit recognition models.The results of the experiment indicate that the facial action unit recognition model can recognize the facial action unit numbers associated with different emotions,and that incorporating prior knowledge of the co-occurrence matrix for action units further improves recognition rates.The high recognition rate of the facial action unit model indicates that the high-level semantic information extracted from facial expressions during facial expression recognition can be used to improve facial action unit recognition,demonstrating the effectiveness,accuracy,and transferability of this approach in extracting facial expression semantic information.(2)In the semantic communication network for face expression recognition task,this paper proposes different semantic communication models for face expression semantic information extraction and encoding for each of the two receiver-side processing methods.In the case that the face image needs to be reconstructed for subsequent tasks at the receiver side,this paper combines the self-encoder in deep learning with the semantic communication network and proposes a semantic communication model based on the self-encoder for face expression recognition,and improves it by adding noise reduction training on the basis of this model.The experimental results show that the quality of face image reconstruction and face expression recognition rate of the semantic communication scheme are higher than those of the traditional communication scheme under various channel conditions.In the case that the receiver does not need to reconstruct the face image,but only needs to recognize the corresponding expression of the face image to perform the corresponding task,a semantic information extraction and coding algorithm is proposed to decouple the semantic information of identity and expression in the expression image,and the extracted semantic feature information related to the expression is applied to the semantic communication network model for the face expression recognition task.The experimental results show that this model has better face expression recognition accuracy than other schemes in the case of poor channel conditions.