Real Time Face classification and Shape based feature extraction with Deep Neural networks

Feature fusion and pose detection with real time face images is based on synergy techniques used in HyperFace and occlusion-based techniques than employ L-NMS or IRP. We implement real time categorizing of the live images by Head pose estimation into straight and side face, these images are separately trained over two Neural network models executing in parallel and then extracting certain set of features from the face with shape-based feature fusion techniques using CNN to identify people in a certain occlusion-based environment. Our work enhances the accuracy and speed of detection of faces in the real time scenarios by considering the modified form of HyperFace architecture. Feature based techniques have shown more accuracy in detection of images compared to other face detection methods. Data augmentation is performed over face images to store face features accurately with face images in the database. Face features are predicted based on visibility mentioned during the training of modified Hyperface architecture. The features are thus used to create shapes and find the required angles using law of cosines. Images of people and angles are stored in the database for person identification.