Translator Disclaimer
Chapter 5:
Nonlinear Image Processing Algorithms
A key motivation for the study of nonlinear digital processing is the logical structure of the computer. All algorithms must be reduced to data and instruction flows through logic circuits and this requirement restricts the mathematical structures that are directly appropriate to digital processing. Linear filters involve addition and multiplication. Multiplications, when used in a linear convolution, also often involve real numbers (referred to as floating point numbers in computer architecture parlance), not just integers. Both addition and multiplication of floating point numbers take significantly longer than their integer counterparts since these operations must be implemented via a firmware subroutine involving multiple integer shifts, adds, and multiplies. Although some arithmetic logic units (ALUs) support floating point operations directly in microcode and coprocessors are sometimes added to facilitate floating point operations, these operations are still slower than integer operations.
Online access to SPIE eBooks is limited to subscribing institutions.

Back to Top