Deep Inception Based Convolutional Neural Network Model for Facial Key-Points Detection

The identification of facial key-points is a difficult challenge in computer vision. Each person’s face has incredibly distinctive facial characteristics. The centres and borders of the eyes, the brows, the nose, and the lips are some of the key elements of the face. Thus, finding the facial key features in a particular face is the major aim of the facial key point detection process. This is a very challenging task. In this work, the initial step of the proposed methodology for identifying facial key-points includes a range of training data augmentations to improve the number of training instances and the generalizability of the proposed model. In the second phase, an Inception-based deep neural network model is proposed that accomplishes this task quicker by reducing the training time. While testing, we combined the images from different data augmentation approaches to enable averaged predictions. We evaluate the model design with and without data augmentation techniques based on their Mean Squared Error (MSE) scores on the test set. The research demonstrates that the MSE of the proposed Inception-based model is very low as compared to the other state-of-the-art methods.