4.10 - Video Segmentation

Abstract

This chapter provides an overview of video segmentation methods ranging from simple shot boundary detection and change detection techniques to the more sophisticated motion segmentation, interactive video object segmentation, and tracking methods. Temporal video segmentation methods are discussed in the chapter, followed by change detection methods where both the two-flame method and the methods with memory are covered. Motion segmentation methods are also discussed with an introduction to a dominant motion approach, which aims to label independently moving regions sequentially. The chapter discusses the alternative multiple motion segmentation approach, including clustering in the motion parameter space, maximum likelihood segmentation, maximum a posteriori probability segmentation, and the region labeling methods. The simultaneous optical flow estimation and segmentation method is also addressed because the accuracy of segmentation results depends on the accuracy of the estimated motion field and vice versa. This chapter also deals with semantically meaningful object segmentation with an emphasis on chroma keying and semi-automatic (interactive) object-tracking methods.