Application of the Gestalt principles to the detection of good continuations and corners in image level lines

Abstract

In this paper, we propose an algorithm to automatically detect regular subcurves in a set of digital curves. The decision criterion is based on Helmholtz Principle introduced by Desolneux, Moisan and Morel and formulated in terms of number of false alarms. We apply our algorithm to low-level computer vision. Following Gestalt Theory, good continuation is indeed, one of the most important grouping laws entering into the early perception of objects. We check this on the level lines of images, which give a contrast invariant representation of images. The result is that most objects are good continuations. This experimentally shows that contrast invariance is a very sound hypothesis for low level vision and that objects can be detected independently of contrast. The parameters of the method may be reduced to a single one – the number of false alarms – and we can show that the detection has a very weak dependency on this number.