Table 1.

Performance comparison for the three approaches on images with increasing numbers of objects. A full iteration is comprised of six iterations of level set evolution followed by reinitialization to ensure the stability of all methods. The first of these 30 iterations is reported as the iteration count for convergence. “DNC” indicates that the algorithm did not converge by the 5000th iteration. The misclassification rate counts gaps and overlaps as misclassified, and is computed after convergence.

	Objects	MP	CLS	MGDM
Convergence (full iterations)	4	223	105	33
	8	144	762	87
	16	DNC	459	142
	32	DNC	457	186
Memory usage (average, MB)	4	19	21	11
	8	22	29	11
	16	23	46	11
	32	26	78	11
Computation Time (average per iteration, ms)	4	54	236	51*
	8	81	593	32
	16	183	1143	34
	32	596	2840	35
Misclassification rate (%)	4	0.17	7.63E-4	1.91E-3
	8	0.93	7.63E-4	4.20E-3
	16	4.58	3.05E-3	4.65E-2
	32	14.3	8.01E-3	2.06E-2