Fast and accurate edge-based segmentation with no contour smoothing in 2-D real images

Abstract

In this paper we propose an edge-based segmentation algorithm built on a new type of active contour which is fast, has a low computational complexity and does not introduce unwanted smoothing on the retrieved contours. The contours are always returned as closed chains of points, resulting in a very useful base for subsequent shape representation techniques.

An efficient parameterless quadrilateral-based image segmentation method

“…Some other boundary-based segmentation techniques, as opposed to the aforementioned ones, take the contour model [30], [31], [32], [33], [34], [35] approach in which it always guarantees the formation of closed boundary without the need of any boundary connection tools. In essence, these segmentation techniques start from an initial curve, which is then deformed and split successively according to some predefined models and rules.…”

“…As a result, many of these segmentation techniques require a proper initialization of the initial curve, which is usually chosen manually, in the vicinity of the expected boundaries in order to reduce the computation required in guiding the curve to the expected boundaries and to avoid being trapped in local minima. In [30], the points on the initial curve are regarded as independent cells in living tissue, strictly connected at the same time, which can reproduce, move and die according to their local environment (edge information). Under this model, the deformation of the curve follows a very simple set of rules, without the problems encountered in the energy minimization approach.…”

Generic colour image segmentation via multi-stage region merging

“…Edge-based methods [4,5], on the other hand, find region boundaries by applying edge detection mechanisms, and are limited due to the high number of edges found and by the need to have an effective mechanism to close edges and form contained regions. Histogram-based methods [6,7] analyse peaks in dominant colours in order to establish cluster centers to which pixels are assigned.…”

A Real-Time Solution to the Image Segmentation Problem: CNN-Movels

“…These algorithms generally seem to suffer from an intrinsic high computational complexity and from an effect of contours smoothing which can be undesired. In (Iannizzotto & Vita, 1996) and, later, in (Iannizzotto & Vita, 2000), a new kind of active contour was introduced: this is composed by a chain of autonomous agents (MOVing elements: MOVels), which move independently but in a collaborative fashion over the image, according to some very simple rules and some image features. The idea of exploiting both homogeneity and non-homogeneity as pixel feature for image segmentation appears very attractive to overcome (at least, partly) the problem of noise sensitivity.…”