Chan–Vese Reformulation for Selective Image Segmentation

Abstract

Selective segmentation involves incorporating user input to partition an image into foreground and background, by discriminating between objects of a similar type. Typically, such methods involve introducing additional constraints to generic segmentation approaches. However, we show that this is often inconsistent with respect to common assumptions about the image. The proposed method introduces a new fitting term that is more useful in practice than the Chan–Vese framework. In particular, the idea is to define a term that allows for the background to consist of multiple regions of inhomogeneity. We provide comparative experimental results to alternative approaches to demonstrate the advantages of the proposed method, broadening the possible application of these methods.

Reproducible kernel Hilbert space based global and local image segmentation

“…The particular object (or objects) of interest are indicated by a marker set M, which is a set of points that lie in the object, typically prescribed by the user. We briefly review the work in [27], which achieved local segmentation by using a geodesic distance constraint from [26] (similarly to [28] which used Euclidean distance), and proposed reformulated Chan-Vese like fitting terms more suitable for local segmentation.…”

“…Geodesic distance. Compared to Euclidean distance used for local segmentation [28], the geodesic distance is found to perform better in [27] because it is edge aware. This indicates that edge enhancement can make a big impact through geodesic distance.…”

“…where D G (x) is computed using (47) and (4) and g(Ψβ) = 1 1+ι|Ψβ| 2 . In [27], Roberts and Spencer find the associated Euler-Lagrange equations to solve their model, using a modified AOS algorithm first introduced in [28] to solve the resulting time dependent PDE. Instead of taking that approach, we use the dual formulation method of Chambolle [11], first applied to segmentation problems in the convex relaxed setting in [8].…”

“…More recently, Roberts and Chen [26] proposed to change the Euclidean distance constraint from [28] to an edge weighted geodesic distance constraint. Moreover, in [27] Roberts and Spencer introduced a reformulated Chan-Vese fitting term for selective segmentation and making use of the geodesic distance.…”

Using Deep Image Prior to Assist Variational Selective Segmentation Deep Learning Algorithms

“…The variational approach has been particularly successful in tackling a wide range of problems [1][2][3] without the need for large training sets or time consuming manual labels. A less studied variant is selective variational image segmentation, [4][5][6] which is the task of segmenting a particular object or objects, usually indicated by a set of marker points M input by the user. These marker points are typically placed inside the region of interest, and geometric constraints can be designed to penalise segmenting regions away from the intended object.…”