International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 09 Issue: 08 | Aug 2022
p-ISSN: 2395-0072
www.irjet.net
3D Face Modelling of Human Faces using GANs Varanasi L V S K B Kasyap1, Manikonda V Srikar Janardhan Rao1 1School
of Computer Science and Engineering, VIT-AP University, India ---------------------------------------------------------------------***--------------------------------------------------------------------2. Area Measurement Abstract - 3D Face modelling is not same as 2D Face image generation using DeepFake. This paper suggests a model, in solving the problem of responsive 3D face generation using less training data. By using Deep Convolutional Neural Networks (CNNs), the loss function is defined on feature maps. Optimization problem is solved using Stochastic Gradient Descent (SGD). Generative Adversarial Networks (GANs) are used here to generate 3D Face Model from feature maps. Recurrent Neural Network (RNN) makes it to classify the image to be progressed or not. This model is evaluated against dataset generated with 30 people in laboratory and validates the acceptable performance and boosts up the Inception Score (IS) in 3D Face generation with contemplate limits 3D Face modelling is not same as 2D Face image generation using DeepFake. This paper suggests a model, in solving the problem of responsive 3D face generation using less training data. By using Deep Convolutional Neural Networks (CNNs), the loss function is defined on feature maps. Optimization problem is solved using Stochastic Gradient Descent (SGD). Generative Adversarial Networks (GANs) are used here to generate 3D Face Model from feature maps.
The distance between the features of the face can be calculated using line Euclidean distance between eyeforehead, eye-eye, eye-nose, nose-mouth, eye-mouth, eye-ear. The face can be divided into smaller regions and calculating the area of these small regions and storing them in the database help to generate 3D Face model, however finding the regions of face basing on only one projection is not enough to make 3D Face model, since the other side of the facial data is lost. To overcome this problem, model at least needs three images of same person i.e., Front View, Left Side View, Right Side View and divided as ten regions, six regions, six regions simultaneously. The ten regions of the face for Front view Projection are FR1: the area of the triangle with left eye, glabella(lateral point on forehead)and left ear as vertices, FR2: the area of the triangle with left eye, right eye, glabella as vertices, FR3: the area of the triangle with right eye, glabella and right ear as vertices, FR4: the area of the triangle with left eye, nose and left ear as vertices, FR5: the area of the triangle with left eye, right eye and nose as vertices, FR6: the area of the triangle with right eye, nose and right ear as vertices, FR7: the area of the triangle with left ear, nose and left end point of mouth as vertices, FR8: the area of the quadrilateral with left end point of mouth , nose, right end point of mouth and upper end point of upper lip as vertices, FR9: the area of the triangle with right end point of mouth, nose and right ear as vertices, FR10: the area of the triangle with left end point of mouth ,upper end point of upper lip ,right end point of mouth and right end point of mouth as vertices. The six regions of the face for Left Side View Projection are LS1: the area of the triangle with left eye, lateral point of head and Imaginary Point1(shown in Figure3), LS2: the area of the quadrilateral with left eye, nose, Imaginary Point1,Imaginary Point2 (shown in Figure3) and nose, LS3: the area of the triangle with nose ,mouth and Imaginary Point2, LS4: the area of the triangle with left ear, lateral point of head and Imaginary Point1, LS5: the area of the triangle with left ear, Imaginary Point1,Imaginary Point2, LS6: the area of the triangle with left ear, mouth, Imaginary Point2.
Key Words: Face Modelling, GANs, Feature Modelling. 1.INTRODUCTION Generating 3D Faces from images of 2D Faces is the predominant application of the recent Generative Neural Networks. Besides generating the virtual 3D Faces, features of the face to be generated are obtained using CNNs and by training them over RNN gives the better result. Generating faces using Conditional Generative Adversarial Networks(cGANs) [8], makes the face more realistic. So far, CNNs are used in semantic segmentation, 2D image generation [2], Mu Li et al. developed a method to embed the Human-Identity in CNN. However, the combination of GANs and CNNs could be able to generate 3D Faces with less training data (images, video clips). Yu Song et al. developed Face Recognition Algorithm using facial feature data extraction [1], follows the Euclidean distance measure, Angel Feature measure, Curvature Distance measure and Volume Feature measure. Cahit et al. work of designing a Face Recognition system [3] is used in this paper to recognize the primary facial features (eyes, nose, ear, mouth, forehead). Facial Features play key role in identifying a person. Human Brain also uses these features in recognizing human faces and this spatial data is stored in the synapses, the work of Y.
© 2022, IRJET
|
Impact Factor value: 7.529
3. Identity Loss For transferring the attributes of 2D Faces into 3D model, CNN is used to develop a feature vector, the 2D Face embeddings are enumerated using encoders presented in [9]. Encoder maps images and features to parallel embedding space such that all the features essential for 3D Face model are mapped to feature vector with a high inner product. Since
|
ISO 9001:2008 Certified Journal
|
Page 237