Self-Organizing Maps and Their Applications in Image Processing, Information Organization, and Retrieval - Nonlinear Signal and Image Processing: Theory, Methods and Applications - page 432

(a) (b)

(c) (d)

FIGURE 11.11
Color image quantization. (a) Original image. (b) Quantized color image by the split-merge SOM.
(c) Quantized color image by the two-layer SOM. (d) Quantized color image by the standard SOM
that is initialized as the LBG algorithm.

11.7.4.1 Color Image Quantization
Figure 11.11a shows an RGB color image of dimensions 256 256 with 24

bits per pixel. We compare the performance of single-layer and two-layer
split-merge SOM algorithms to that of the standard SOM in quantizing the
color image of Figure 11.11a with approximately 256 quantization levels. In
our study we use as a figure of merit the MSE given by Equation 11.92, or
equivalently the peak signal-to-noise ratio (PSNR), at the end of the recall
phase. PSNR is defined as follows:

×

10 log 255 2
MSE

PSNR = (in dB). (11.112)

We also quote the number of iterations for each algorithm. Although the
number of iterations is not an objective measure, it is used as a rough estimate