Intelligent coding of facial expression using neural network and morphing

摘要

In face-to-face communication, facial expressions provide valuable information. Video telephones enable such communication between people at distant places. However, in order to obtain a high-quality image, the transmission bit rate must be high. We propose the basic principle for intelligent coding of facial expressions, and suggest the possibility of an extremely low-bit-rate communication system using the code. This system encodes only an image of a speaker's face and greatly reduces the transmission bit rate (about 500 bits/sec) using knowledge of the facial expressions of the speaker. Natural facial expression and their changes are observed through experiments.