Abstract
Facial recognition is a process where we can identify or verify a person from digital image or videos and is used in ID verification services , protecting law enforcement ,preventing retail crime etc. Past work on automatic analysis of facial expression focuses on detecting the facial expression and exploiting the dependencies among AU’s. But, spontaneous detection of facial expression depending on various factors such as shape, appearance and dynamics is very difficult. Joint learning of shape , appearance and dynamics is done by a deep learning technique.This includes a convolutional neural networks and bidirectional long short term memory(CNN-BLSTM). This combination of CNN-BLSTM excels the modeling of temporal information. FERA2015 dataset achieves the state of art.
Highlights
One of the strongest indications for emotion is our face
Ekman and Friesen promotethe Facial Action Coding System(FACS) provides a scientific and comprehensive system, by in lieu of bycombination of individual muscle actions mentioned as Action Units (AU)
The co-effects caused by co-occurring Action Units is a factor, which increases the difficulty level in training the data
Summary
Most of the muscles are triggered by one single nerve called facial nerve. Ekman and Friesen promotethe Facial Action Coding System(FACS) provides a scientific and comprehensive system, by in lieu of bycombination of individual muscle actions mentioned as Action Units (AU). The co-effects caused by co-occurring Action Units is a factor , which increases the difficulty level in training the data. Facial features are classified intoappearance feature and geometric based features. Appearance based feature captures local and global appearance changes, for example Harr feature Local binary pattern, Gabor wavelets, Canonical appearance. Geometric feature based includes direction or magnitude of skin surface and salient feature points. Quick access to the data, increased computing power and high performance are the three factors which increases the accuracy
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have