Improvement of Face Quality using Model-Based Face Region Extraction and Region-Based Quantization

Abstract

This paper presents a model-based region extraction method and demonstrates that ITU-TH.263+ with region-based quantization improves picture quality of face area. Our scheme has resulted in improvement of objective quality of face regions in Y-SNR by around 1dB at 9.6kbps, and about 2dB at 16kbps, as compared with H.263+ ONLY.