The perception of the image world
Brian Funt, Kalpna Gupta, Louis. Brassard · 1999
To understand vision one first has to understand images and to understand images one needs to assume that images are structured. This is a theoretical position similar to the one hold by J. J. Gibson. To sustain this position one has to provide evidence to support the claim that images are structured and this is the main objective of this thesis. For that one need a geometrical framework for the expression of a large class of image regularities. The concept of “crease structure” provides such a framework. Crease structures are essential geometrical features of the Gaussian scale-space field and of the scale-space-time field of an image. They are one-dimensional mathematical entities which capture global structures in images. This concept arose from the observation that the crease networks of a large varieties of static and dynamic images have sub-graphs exhibiting regular crease patterns. More than twenty of these crease structures are identified using a specially designed crease network computation and visualization algorithm. Instances of crease structures are shown into a variety of images from natural scenes, abstract sketches, to well known images used in experimental psychology. Empirical evidence for causal relations between the presence of these crease structures and the human visual perception of simple curves and shapes, grouping, perception of bilateral symmetry axes, perception of movement in the tunnel effect and many more are provided. A method for classifying crease structures into a structural hierarchy tree based on their scale-space structural evolution is described.