Eye and lip enhancement for low-rate videophony

A generic parametric coding enhancement for employment in very low-rate videophone systems is proposed in order to improve the subjectively important eye and lip representation. The four-step algorithm 1 identifies the position of the eyes, nostrils, lips and nose using pre-precessing and pattern-matching and after luminance adjustment and smoothing it overlays the best-matching entries from a trained codebook, if deemed beneficial in image quality terms. The benefits of the technique are demonstrated using the example of an 11.36 kbps fixed-rate quad-tree codec.