Input image sequences of video conferencing are captured from video camera directly. Therefore quality of video camera has an influence on video compression. As video camera itself has capabilities of brightness control and automatic gain control, brightness of image sequences from video camera changes depends on the changes of lighting condition. When a video camera is sensitive to small changes, the scene from the camera looks brighter or darker by slow degrees. Although these variances of luminance level are negligible to the human visual system, they disturb the motion estimation function in the block based video encoder. Therefore poor motion estimation results in the decrease of compression ratio, worse video quality or less fps (frame per second). In low bit-rate videoconference this problem can be amplified because more frame skip makes difference of brightness between two frames clear.
In this thesis, I propose a pre-filtering scheme to keep up the luminance level within the limit of small variances. The influence of small brightness change to video encoder is tested on H.263 encoder and general noise reduction filters and proposed pre-filter. Simulation results show that general noise filter cannot keep up the stable luminance level whereas the pre-filter on image sequences with brightness variance achieves significant improvements.
