Fig. 5.

The relative speedup performance of DCBFC to the other methods (using $R_t$ as the similarity matrix). Data were tested on three types of image resolutions: $64\times 64$, $128\times 128$, and $256\times 256$. (a) $11\text{-}\text{functional modules}$ type data with 5 dB SNR level. (b) $11\text{-}\text{functional modules}$ type data with $-10\mathrm{dB}$ SNR level. (c) $50\text{-}\text{functional modules}$ type data with 5 dB SNR level. (d) $50\text{-}\text{functional modules}$ type data with $-10\mathrm{dB}$ SNR level. Compared with the other methods (k-means, hierarchical, spectral, PCA-k-means, PCA-hierarchical, and spectral-threshold clustering), DCBFC, respectively, provides a 3.7 to $\mathrm{16,542}\times$ performance boost over k-means clustering, a significant 211.9 to $2855\times$ performance boost versus hierarchical clustering and almost 1.1 to $133.2\times$ performance boost versus spectral clustering. Compared with k-means and hierarchical clustering using PCA, DCBFC also has 2.0 to $25.5\times$ and 41.7 to $2171\times$ performance boost. And has 1.3 $-330.4\times$ performance gain versus spectral-threshold clustering.