||Directional Filter Bank(DFB) can be thought of as a bank of
filters which decomposes the Fourier Transform of an Image into Wedge-shaped
like pass-bands. In the spatial domain, these pass bands correspond to features
(not edges) of an image in a particular direction.
DFB was originally designed for compression purpose, but later on it was
used in different image processing application like image enhancement,
fingerprint recognition, texture analysis, etc,.
DFB is generally implemented in binary tree-structured way, where each node
(image) decomposes into two further child nodes (sub-bands). The directional
information present in the parent node is divided into two directionally
orthogonal sub-bands. The sum of pixels in both the sub-bands is equal in
size to their parent node. So, in DFB, directional resolution is inversely
related to spatial resolution, and due to the presence of decimators, it
cannot be kept under the category of TI systems. One main drawback of the
DFB was the fact that due to non-diagonalization of overall downsampling
matrix, all the outputs after the second stage were visually distorted.
Later on this drawback was removed by the use of a Visualization matrix.
We propose improved directional filter bank which can be used as an LTI System.
The major feature of the proposed system is the fact that it doesn’t
require any visualization matrix at any stage and directional resolution is
made independent of spatial resolution. It requires no
interpolation during the enhancement process as it was mandatory step in the
previously proposed DFB.