Abstract: Semantic object segmentation is a key technique for many content-based multimedia applications intensively investigated in the last decade. From the application viewpoint, three classes of practical semantic object segmentation approaches are discussed. Specifically, these approaches include realtime automatic object segmentation, non-realtime interactive object segmentation, a prior knowledge based object segmentation. Having analyzed the advantages and drawbacks of each class, we conclude that high-level semantics should be integrated in the semantic object segmentation approaches to further expand its applicability. We also give a preview of the emerging visual attention based semantic object segmentation approaches.

Key words: automatic object segmentation, high-level semantics, interactive object segmentation, prior knowledge, visual attention
, semantic object segmentation