The Non‐Accidentalness Principle for Visual Perception
Agnès Desolneux, Lionel Moisan, Jean‐Michel Morel · 2013
To analyze images automatically with a computer the way a human being would do it, it first needs to answer, from a computational viewpoint, the question: “How can we go from qualitative to quantitative?”. Such a goal can be achieved by combining two principles of visual perception: the non-accidentalness principle and the Gestalt grouping laws. This chapter details these two principles, and shows how they can be combined in the framework of the so-called a contrario methodology. It illustrates this methodology on three examples: alignments, contrasted curves and good continuations. The chapter ends with an experimental section in which it proposes some protocols to check the validity of the a contrario methodology.