| In today’s society,the human’s is not only served to express different emotional but also used for videos,medical diagnosis security system.Among these domains,facial expression synthesis and 3D facial reconstruction are currently key research areas.Traditional facial expression generation algorithms typically rely on the positional changes of facial landmarks to synthesize facial expressions.However,this approach fails to capture the subtle variations in facial muscles,then it is short of naturalness and realism.Regarding 3D facial reconstruction,most traditional algorithms are based on generic 3D facial models.However,due to the inherent structural differences in individual facial features,will have no high accuracy and characteristic,if the 3D fail model is used in all the cases.In response to the above issues,this paper focuses on the following research topics:For the problems of blurry,unnatural,and unrealistic facial expressions in traditional facial expression synthesis algorithms,this paper proposes a facial expression synthesis algorithm based on normalization and attention mechanism in a Generative Adversarial Network(GAN)framework,named GCT-Star GAN.Firstly,the Star GAN algorithm is employed to extract the feature vector of the original facial expression image.Secondly the extracted facial expression feature vector,along with the target expression label vector,is passed through a threshold channel control unit composed of normalization and attention mechanisms.The feature vector is normalized using L2 normalization to avoid local confusion and overfitting.The attention mechanism assigns higher weights to facial expression features,enabling the extraction of finer-grained information and better preservation of facial expression characteristics.Final this results in more realistic and natural synthesized facial expressions.Experimental comparisons are conducted to validate the effectiveness of the proposed algorithm.To solve the problem of low accuracy and indistinctive features in traditional 3D facial reconstruction algorithms,this paper utilizes the Media Pipe Face Mesh algorithm for 3D facial reconstruction.The Media Pipe Face Mesh algorithm generates corresponding 3D facial landmarks based on specific facial expression images.The 3D facial reconstruction is performed using texture mapping based on these 3D facial landmarks.In comparison to traditional generic 3D facial models,the algorithm employed in this paper achieves more accurate and realistic 3D facial reconstruction results.To address the aforementioned issues and meet users’ applicational needs,this thesis designs a system for facial expression synthesis and 3D reconstruction.If thesis consists of user module,file upload module,facial expression synthesis module,and3 D reconstruction module.To achieve the functionalities of facial expression synthesis and 3D reconstruction,the system is based on a B/S architecture and utilizes front and back ends separation techniques.The system interface is implemented using the QT framework,while the back-end programs are developed on the Py Torch framework by the Python language.Through functional and performance testing,the system is validated to meet the design requirements and achieve the intended design goals. |
