Extraction of High-Resolution Face Image from Low-Resolution and Variant Illumination Video Sequences

In video surveillance, resolution and illumination are two main factors affecting the performance of face detection and recognition. In this paper, a new framework is proposed to extract face images from low resolution and variant illumination video sequences, which includes two parts: face detection and face hallucination. We create skin GMMs in CbCr and H-SV space to detect face in different low resolutions and lighting conditions. Next, eigentransformation by 2DPCA is used to face hallucination, and produce a global high-resolution face. Furthermore, we adopt Coupled PCA method to obtain facial local residue. Then global- and local-face are combined into a high equality face image. Experimental results show that we can obtain high resolution face images for recognition.