On the Relationship between Variational Level Set-Based and SOM-Based Active Contours

Table 1

A summary of the Active Contour Models (ACMs) reviewed in the paper.

ACM	Reference	Regional information		Main strengths/advantages	Main limitations/disadvantages
		Local	Global
GAC	[24]	No	No	Makes use of boundary information.	Hardly converges in the presence of ill-defined boundaries.
				Identifies accurately well-defined boundaries.	Very sensitive to the contour initialization.
CV	[26]	No	Yes	Can handle objects with blurred boundaries in a global way.	Makes strong statistical assumptions.
				Can handle noisy objects.	Only suitable for Gaussian intensity distributions of the subsets.
SBGFRLS	[68]	No	Yes	Very efficient computationally, and robust to the contour initialization.	Makes strong statistical assumptions.
				Gives efficient and effective solutions compared to CV and GAC.	It is hard to adjust its parameters.
LBF	[71]	Yes	No	Can handle complex distributions with inhomogeneities.	Computationally expensive.
				Can handle foreground/background intensity overlap.	Very sensitive to the contour initialization.
LIF	[32]	Yes	No	Behaves likewise LBF, but is computationally more efficient.	Very sensitive to noise and contour initialization.
LRCV	[31]	Yes	No	Computationally very efficient compared to LBF and LIF.	Very sensitive to noise and contour initialization.
LSACM	[73]	Yes	No	Robust to the initial contour.	Computationally expensive.
				Can handle complex distributions with inhomogeneities.	Relies on a probabilistic model.
GMM-AC	[75]	No	Yes	Exploits prior knowledge.	Makes strong statistical assumptions.
				Very efficient and effective.	Requires a huge amount of supervised information.
SISOM	[38]	No	No	Localizes the salient contours using a SOM.	Topological changes cannot be handled.
				No statistical assumptions are required.	Computationally expensive and sensitive to parameters.
TASOM	[39]	No	No	Adjusts automatically the number of SOM neurons.	No topological changes can be handled.
				Less sensitive to the model parameters compared to SISOM.	Sensitive to noise and blurred boundaries.
BSOM	[93]	No	Yes	Exploits regional information.	Topological changes cannot be handled.
				Deals better with ill-defined boundaries compared to SISOM and TASOM.	Computationally expensive and produces discontinuities.
eBSOM	[94]	No	Yes	Produces smooth contours.	Topological changes cannot be handled.
				Controls the smoothness of the detected contour better than BSOM.	Computationally expensive.
FTA-SOM	[96]	No	Yes	Converges quickly.	Topological changes cannot be handled.
				Is more efficient than SISOM, TASOM, and eBSOM.	Sensitive to noise.
CFBL-SOM	[97]	No	Yes	Exploits prior knowledge.	Topological changes cannot be handled.
				Deals well with supervised information.	Sensitive to the contour initialization.
CAM-SOM	[98]	No	Yes	Can handle objects with concavities, small computational cost.	Topological changes cannot be handled.
				More efficient than FTA-SOM.	High computational cost compared to level set-based ACMs.
CSOM-CV	[102]	No	Yes	Very robust to the noise.	Supervised information is required.
				Requires a small amount of supervised information.	Suitable only for handling images in a global way.
SOAC	[103]	Yes	No	Can handle complex images in a local and supervised way.	Supervised information is required.
				Can handle inhomogeneities and foreground/background intensity overlap.	Sensitive to the contour initialization.
SOMCV	[104]	No	Yes	Reduces the intervention of the user.	Is easily trapped into local minima.
				Can handle multimodal intensity distributions.	Deals with images in a global way.
SOM-RAC	[105]	Yes	Yes	Robust to noise, scene changes, and inhomogeneities.	Very expensive computationally.
				Robust to the contour initialization.