Segmentation and composition are important processes in video and movie production and are rapidly developing research areas. To segment a source object and seamlessly blend it into a target scene satisfying uniformity (homogeneity) criterion which is based on feature(s) derived from spatial or/and temporal information, a large number of algorithms have been proposed. To alleviate problems inherent to earlier efforts on automatic segmentation and composition, much work has been performed in an interactive way, which requires user’s assistance to guide the process....[ Read more ]

Segmentation and composition are important processes in video and movie production and are rapidly developing research areas. To segment a source object and seamlessly blend it into a target scene satisfying uniformity (homogeneity) criterion which is based on feature(s) derived from spatial or/and temporal information, a large number of algorithms have been proposed. To alleviate problems inherent to earlier efforts on automatic segmentation and composition, much work has been performed in an interactive way, which requires user’s assistance to guide the process.

In this thesis, we will first review several representative and recent algorithms in the area of segmentation and composition. Then our algorithm for segmenting and compositing video object with simple user interaction will be described. To generate a seamless video composite, we propose an approach that integrates both the segmentation and composition processes, by taking into consideration both the source and target videos. We call this target sensitive video segmentation for video compositing. We also demonstrate impressive results and comparisons to evaluate the performance of our algorithm.