Multiple-face tracking system for general region-of-interest video coding

Abstract

An algorithm has been developed which locates and tracks faces in color video sequences for rise in region-of-interest video compression. Our face-tracking algorithm detects faces in the hue-saturation-value (HSV) color space and reduces false alarms based on various cues, including size, shape position, and aspect ratio. The algorithm also performs temporal filtering to enforce consistency from frame to frame. The face tracker is integrated in the rate control of a video encoder, which allows more bits to be committed to the face regions at the cost of reduced bits in the non-face regions. Thus, image quality can be greatly improved in face regions at the cost of reduced quality in background regions, resulting in better overall subjective quality for many sequences.