Post inserted object calibration for stereo video rectification

Stereo rectification removes vertical disparity between stereo image pair and it can improve visual quality of stereoscopic displays. This work addresses a practical issue when a stereo video input has been enhanced with post inserted objects (PIOs) such as captions and graphic objects. In such cases, previous stereo rectification methods would result in errors in the PIO regions due to PIOs' artificially edited epipolar geometry than those of the scene objects. To avoid such problem, this work proposes a PIO calibration method, which involves extracting PIOs, filling the PIO pixels, and re-inserting the PIOs into the rectified stereo video. The method constructs PIO models to extract PIOs in real stereo videos. Also, a modified stereo rectification procedure is proposed to work with the PIO calibration method, which reduces computational load and avoids video jitter artifacts. The method is tested on both real and simulated stereo videos to verify its validity.