A comparative review of boundary completion problem

The boundary completion problem plays a seemingly indispensable role in computer vision and image understanding research. We define boundary completion as growing or extracting from an image the salient structure or form that imparts an informative semantics. Its objectives and scopes are discussed. Literature survey results are contrasted and commented on the basis of type of input data, insight, prior assumption and solution techniques. Finally, a contrast-based attributes theory, related to relaxation labelling and Markov random field, is outlined as a feasible generic solution to this problem.