The generator creates for each pixel in the image the source address from
which a block of 4 pixels is to be derived along with the contrast and
brightness adjustment to be applied to them to derive the appropriate similarity
TeX
The purpose of this class is to provide a fast means of
indexing blocks of 4 pixels, (2x2) blocks, so as to determine
which block of a given larger size is most similar to a particular
block of a smaller size.
TeX \subsection{Methods Inherited from \em Subscriptable}
Subscription by a scalar returns the appropriate pixel
in the image, allowing for the offset of this image from
the start of the image data.
TeX
fSubscription by a scalar returns the appropriate pixel
in the 0th plane of the image, allowing for the offset of this image from
the start of the image data.
TeX
fSubscription by a scalar returns the appropriate pixel
in the 0th plane of the image, allowing for the offset of this image from
the start of the image data.
TeX this returns a vector of integers containing the bext match as follows
\begin{tabular}{|c|c|c|c|c|c|}\hline
plane&x&y&rotation&shift&contrast\\
\hline
\end{tabular}
where
plane,x,y indicate the source from which the matching patch can be found;
rotation indicates how many 90 degree clockwise rotations are needed to
bring the source patch into the same orientation as the target;
shift and contrast are used to perform an affine transform in the brightness
space of the source patch to make it as close as possible to the target patch.
TeX This method allows the depth as well as area of an image
to be altered if it is reduced the planes are aggregated
if increased they are interpolated
TeX
given a rotated target vector in least first form, the rotation applied to get it to that
form, and a row from the table this derives the optimal transform containing:
\begin{tabular}{|c|c|c|c|c|c|}\hline
plane&x&y&rotation&a&b\\
\hline
\end{tabular}
Let the source vector after rotation be $\{p,q,r,s\}$ and the target vector be
$\{i,j,k,l\}$
\begin{description}
\item[plane] derived from the 5th column of the table
\item[x] derived from the 6th column of the table
\item[y] derived from the 7th column of the table
\item[rotation] Let the rotation required to bring the source
into normal form be $r_s$ and let the
rotation required to bring the target into such alignment
be $r_t$ then the rotation required to bring
the source into alignment with the target is
$r_s-r_t$.
First parameter is a hierarchy of vector quantization
tables
Second parameter a pair of tables encoding the x and y
discrepancies appropriate for any pair of codebook entries
Levels specifies the number of halvings in size to be achieved
inputbits specifies the number of bits in the integers used in training
internal bits specifies the accuracy of the integers used internally
outputbits specifies the number of bits to which the vectors will finally be quantized
TeX The ImageTransfer class is designed
to provide a means of passing one plane of data
to the intel image processing library for
treatment with the MMX hardware.
TeX
\def\frac#1#2{(#1)\div(#2)}
The class Jimage\footnote{\copyright Turing Institute, 1998}
is an abstract class to support image processing operations.
TeX one image less than another if all pixels
less than corresponding ones, thus it
can be implemented by injecting $<$ between
the elements of a and b, and then applying
$\times/$ to the result
TeX one image less than another if all pixels
less than corresponding ones, thus it
can be implemented by injecting $<$ between
the elements of a and b, and then applying
$\times/$ to the result
Takes two images and returns x,y disparity images
disparities in fractions of source image dimensions
needed to move the left image points onto corresponding
points on the right image.
TeX
This operator takes an image in RGB and a saturation value b and
maps to the set of basis vectors
$$
\pmatrix{X\cr Y\cr Z}=\pmatrix{
$1\over \sqrt 3$&$1\over \sqrt 3$&$1\over \sqrt 3$\cr
0&$-1\over \sqrt 2$&$1\over \sqrt 2$\cr
$-1\over \sqrt 2$&$0.5\over \sqrt 2$&$0.5\over \sqrt 2$}\pmatrix{R\cr G\cr B}
$$
Matrix chosen to have unit length vectors.
TeX
This operator converts an image in RGB format to an XYZ image
by multiplying by a conversion matrix as follows:
$$
\pmatrix{X\cr Y\cr Z}=\pmatrix{0.412&0.357&0.18\cr
0.212&0.715&0.072\cr
0.019&0.119&0.95}\pmatrix{R\cr G\cr B}
$$
TeX
\subsection*{Internal representation}
Observe that if the pixels are rotated into an orientation such that the
0th pixel is the lowest valued, then by storing an additional parameter,
a brightness shift B, the 0th pixel of each patch can be converted to 0.
TeX
Images dx and dy contain the discrepancy in x and y positions
from which source pixels to be obtained
let $r$ be the result, $s$ the source then
$$r_{i,j}=s_{i+dx_{i,j},j+dy_{i,j}}$$
TeX
This operator converts an image in XYZ format to an RGB image
by multiplying by a conversion matrix as follows:
$$
\pmatrix{R\cr G\cr B}=\pmatrix{3.24&-1.537&-0.498\cr
-0.969&1.875&0.041\cr
0.055&-0.204&1.057}\pmatrix{X\cr Y\cr Z}
$$