Context-based natural image parsing: A critical survey

Image parsing is an essential part of a large number of image understanding systems. It encompasses techniques for segmenting images into meaningful objects and labeling them with semantic classes. Both sub-problems of segmentation and classification are impossible to deal with in most natural domains without implicit or explicit consideration of context provided by other sources (i.e. sensors or a domain knowledge) or emerging from the image itself (neighborhood or global image features). In this paper we will discuss different approaches that exploit the image context in order to achieve reliable parsing of natural scenes.