Research On Interactive Image Segmentation

Image Segmentation is a very important part of image processing. It is wildly used in areasuch as Pattern Recognition, Medical Image Process and Image Editing, and has always beenattractive in research. We studied most of the typical algorithms in image segmentation, andcategorized them based on the way they use the information of image. Since automatic image segmentation is hard to implement in many cases, sometimesinteractive operation is needed. We studied the state of the art interactive image segmentationalgorithms and evaluated their performance. According to the object of segmentation, there aretwo different kind of interactive image segmentation: the hard segmentation and alpha matting.GrowCut is one of the interactive hard segmentation with good performance and extension.It is implemented under the framework of Cellular Automata. We will introduced this algorithmin detail, and improve it through the interactive method and Neighborhood System. After alpha matting process, we may need to refine the result, and new interaction will beadded, but most of these algorithms do not handle this situation. So we set up a new algorithmbased on Cellular Automata. It uses the transparence as its intrinsic property, and achieves theobject through the communication of nearby pixels. This new algorithm can react naturally touser's re-interaction, and give nice performance in many cases.