Facial Feature Detection and Automatic Lip Reading
Posted:
Krish Kabra, Calvin Chang, Amit Mondal and Satvik Anand
Course: CS 269 - Deformable Models in Computer Vision
Instructor: Prof. Demetri Terzopoulos
Quarter: Fall 2020
Abstract
Facial feature detection (FFD) is an important task used in a multitude of applications including facial recognition, animation, expression analysis, and 3D modelling. In this report, we survey 6 well-established FFD methods: active shape models, active appearance models, active contour models (ACM), localized region-based ACM, supervised descent fitters, and ensemble of regression trees. We utilize the LFPW and 300-W IBUG facial databases to train our models, as well as to qualitatively and quantitatively analyze model performance. Finally, we show an example application of FFD to the automatic lip reading task using two deep learning frameworks: a feature-only network and an end-to-end network.