Facial Key points Detection using MobileNetV2 Architecture

With the popularity of social media like Instagram and Snapchat, facial filters or beautifying filters are used more often. These applications cannot store raw images of faces when used every time. Thus, they need unique characteristics of a face to which the filters can be applied. These unique characteristic points of every face are called facial key points. With the increasing use of such applications, facial key point detection has become a popular topic. The objective of key point detection is to extract the coordinates of the unique points in the face which are necessary and sufficient to detect the image. Every person has a different face and the key point coordinate for each is very different from the other. Thus, detecting the key points becomes a difficult task. The detection becomes more challenging depending on the angle with which the image is taken and the light exposure of the image. In this paper, we propose the use of deep learning architecture, MobileNet version2 (MobileNetV2) [26] to solve this complex problem as it proves to be better than the traditional architectures available. Our aim is to detect 15 key points of the given facial image using CNN with MobileNetv2 architecture to obtain lower loss and better accuracy. The baseline model used is a single hidden layer neural network and convolutional the advanced model is a Convolutional Neural Network with MobileNetV2 architecture. The experimental results have shown 84% accuracy as compared to the present state-of-art algorithms.