Topology: A Theory of a Pseudometric-Based Clustering Model and Its Application in Content-Based Image Retrieval

Table 1

Comparing existing clustering paradigms.

Related works
Clustering paradigm	Works	Limitations
Partitional	[1–7]	(i) Only suitable for hyper-spherical separation.; (ii) Optimization problem (non-guaranteed minimum value).
Hierarchical	[1, 3, 8–12]	(i) Suitable for small databases.; (ii) Problem to find optimal number of groups.
Density-Based	[1, 3, 13–16]	(i) Main disadvantage is defining the priori density function.; (ii) Suitable for databases following a given density function.; (iii) Computationally heavy.
Spectral	[1, 3, 21–24]	(i) Frequency domain transformation.; (ii) Preserving time-frequency correspondence.; (iii) Optimization problem (non-guaranteed minimum value).; (iv) Computationally heavy.
Gravitational	[25–29]	(i) Suitable for small databases.; (ii) Only suitable for hyper-spherical separation.; (iii) Optimization problem (non-guaranteed minimum value).; (iv) Computationally heavy.