Study Of Image And Video Segmentation Techniques

In the last decades, the image and video processing technology has been explored widely. Not only the movie and animation industry, but also many amateur users expect to edit the source material captured by themselves. Image and video segmentation techniques play an important role in material editing techniques. The extracted foreground object can be integrated into post-processing work such as matting and compositing, to achieve the purpose of material reusing. So studying on the image and video segmentation techniques and developing a friendly user experienced material editing tools have great significance.This paper focuses on the research of image and video segmentation techniques. In recent years, many excellent image segmentation approaches have been proposed. But along with the development of digital image and video capture devices, many high definition multimedia materials have been produced. The traditional segmentation approaches are incapable to work on these large data volume materials effectively. Multi-resolution segmentation approach reduces the computation cost and improves the performance of the original method, we also add some extra constraints to ensure the correct result during the multilevel segmentation process.In video area, due to the difficulties such as large data volume and spatial-temporal coherence, extracting moving object from a video is a big challenge. We extend the multi-resolution image segmentation approach to video area to reduce the time and space cost of the traditional method. Meanwhile, we propagate the intermediate result at low resolution level to ensure the spatial-temporal coherence of the segmentation result.At last, we proposed a 3D volume based multi-resolution video segmentation system. It provides a good user interface to slice and label the video volume easily. Additionally, we use the 3D volume based segmentation approach to overcome the unreliability caused by the intermediate result propagation between successive frames. The experiment results show that, comparing with other image and video segmentation approaches, our work greatly reduces the time and space cost and can produce the ideal segmentation result.