Human-Content and Gesture-Event Video Coding

Currently, bandwidth limitations pose a major challenge for delivering high-quality multimedia information to users. In this research, we aim to provide a better compression of human-centered video sequences such as lectures, monologues, and presentations. Based on the idea that people pay more attention to face and hand regions in videos containing people speaking, our approach encodes those regions with higher resolution than the remaining image. Using computer vision techniques, we segment and track the subject’s face and hands. The face region is assigned the highest salience value. Gesture analysis of the hands is then used to encode important gesture events at high salience and non-gestures at a lower value. We demonstrate the differential video coder with the production of three highly-salient, low-bandwidth video monologue sequences.