Need of: a) initial contour, b) proper parameters, c) long runtime.

Try it yourself: http://www.creatis.insa-lyon.fr/~bernard/creaseg/

Very nice short summary. One can also add that active contours end in local minima of the optimization function (e.g. piece-wise mumford shah functional). But this is pretty much already covered in a)

see also our recent publication

S. Petroudi, C.P. Loizou, M. Pantziaris, C.S. Pattichis, “Segmentation of the common carotid intima-media complex in ultrasound images using active contours,” IEEE Trans. Biomed. Eng., vol. 59, no. 11, pp. 3060-3069, 2012.

The paper entitled Active Contours Without Edges introduced a model which is not based on edge detection (gradient). In the classical model initial contour should be set correctly, but in Chan-Vese model the initial curve can be placed anywhere in the image.

Although the paper author claim the model can work in salt and paper noise, it can not work correctly. 

In the experimental result section, many of the parameters are generally fixed. the only parameter which hast to be set according to the image is u. 

u is a scaling parameter for the length of the curve. By decreasing u, the length of the curve increases and it makes the model possible to detect smaller objects. Therefore, setting suitable parameter for u is a problem.

Chan-Vese model for active contours solved initial contour and proper parameters limitation to some extent, but long runtime is still there.

The key limitations of the Active Contour method are:

1. It can often get stuck in local minima states; this may be overcome by using simulated annealing techniques, which is the cause of computational complexity.

2. They often overlook minute features in the process of minimizing the energy over the entire path of their contours.

3. Accuracy is governed by the convergence criteria, which is used in the energy minimization technique. Higher accuracy require tighter convergence criteria and hence, longer computation times.

4. When image size is too large, this method works slowly.

5. It is not capable to segment the nearest objects.

6. This method is not so good for video related operations.

1.snakes don't see object boundaries in the image unless it gets very close to it.

2.Cannot follow topological changes of objects