Gaze correction for videoconferencing

• Main Author: Ang, Billy Thiam Guan
• Other Authors: Cham Tat Jen
• Format: Final Year Project
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/42388
• Institution: Nanyang Technological University

A common tool in the area of visual communication would be the web cameras, used for capturing scene images that are to be streamed across the network. Yet, it is a common problem for any web camera to capture images that are offset to the gaze of the user, hence losing the eye-contact with the opposite party. By not maintaining eye contact, visual communication users may experience degrading effects and thus may not enjoy the true benefits from the technology. As such, the Cox’s Algorithm was first introduced to provide a software solution towards generating gaze-corrected images, while Criminisi later proposed an improved 4 planes dynamic programming solution that sets to improve on the Cox’s Algorithm. Both algorithms requires the left and right cameras to provide synchronized and epipolar-rectified images, thus reducing the computation of the stereo correspondences to a 1-dimension search space along the horizontal raster lines of a pair of rectified images. Improvements such as reducing the image resolutions and search space have been proposed in this report. The image resolution reductions set to find a suitable resolution that can help synthesize cyclopean view at a faster rate, while the search space reduction looks into possible way of avoiding computation on a certain portion of the matching graph.