11Linear Transforms

Linear transformation extends and generalises local linear filtering of Eq. (9.4) by removing the restriction of the window. Each output value is a linear combination of all of the input pixel values:

This enables linear transforms to separate different components of an image, for example separating signal from interference or noise.

Direct implementation of Eq. (11.1) is very expensive. For an upper N times upper N input image, each output value requires upper N squared multiplications and additions. Therefore, an upper N times upper N output requires upper N Superscript 4 operations. Many useful transforms are separable in that they can be decomposed into separate, independent, one‐dimensional transforms on the rows and columns. In this case, Eq. (11.1) simplifies to

(11.2)upper Q left-bracket u comma v right-bracket equals sigma-summation Underscript y Endscripts w left-bracket v comma y right-bracket left-parenthesis sigma-summation Underscript x Endscripts w left-bracket u comma x right-bracket upper I left-bracket x comma y right-bracket right-parenthesis comma

reducing the number of operations to 2 upper N cubed. Separable transforms ...

Get Design for Embedded Image Processing on FPGAs, 2nd Edition now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.