The K-SOM has been successfully used in quantizing
digital images. Many hardware realizations of the K-SOM
quantizer have been published in order to speed up part or
all of the operations with similar drawbacks of poor
resulting image quality compared to software counterparts.
Mainly, this resulted from the fact that to make the system
compact and be synthesizable on a single chip FPGA, the
internal data representations and operations within the
hardware platforms are required to be all integer based. In
this paper, we proposed a hardware centric K-SOM quantizer
algorithm which relies on a rational-based representation
of the codebook and learning kernel. This extends
the capability of the quantizer to accept an approximated
non-linear learning kernel. The experimental results proved
that the quality of the outcome images was superior to
previous implementations with an acceptable throughput.
The resource utilizations and frame rate throughput of the
proposed approach were just a bit worse than the predecessor
implementations. It was, however, still possible to
synthesize the whole system onto a single moderate density
FPGA. By following the proposed algorithm, it opens an
opportunity for a hardware quantizer to accept different