17-07-2012, 10:19 AM
Subpixel Image Scaling for Color Matrix Displays
[attachment=27784]
Abstract
The perceived resolution of matrix displays increases
when the relative position of the color subpixels is taken
into account. ‘Subpixel rendering’ algorithms are being
used to convert an input image to subpixel-corrected display
images. This paper deals with the consequences of
the subpixel structure, and the theoretical background
of the resolution gain. We will show that this theory
allows a low-cost implementation in an image scaler.
This leads to high flexibility, allowing different subpixel
arrangements and a simple control over the trade-off
between perceived resolution and color errors.
1 Introduction
To generate full color images, matrix displays like
Plasma Display Panels (PDPs) and Liquid Crystal Displays
(LCDs) use three spatially displaced primary
color subpixels (red, green and blue, RGB) per full
color pixel. These subpixels are arranged in some repeating
pattern, like the ’vertical stripe’ arrangement,
shown in Figure 1.
a b c
Figure 1: In a matrix display (a), each full color pixel
consists of several primary color subpixels (b), here in
the ‘vertical stripe’ arrangement. The question arises
whether these subpixels can give extra resolution when
the grouping into full color pixels is released, as illustrated
by this simple example ©.
Each subpixel is given an intensity corresponding to the
value of the color component at the corresponding full
color pixel location in the image. When the subpixels
michiel.klompenhouwer[at]philips.com
†g.de.haan[at]philips.com
are small enough, they are not individually visible at a
normal viewing distance, so that the viewer will only
perceive the resulting color (tristimulus value) for that
location in the image: so-called ‘color blending’ occurs.
Figure 2 shows the basic diagram of the signal flow in a
matrix display, as it will be discussed in this paper. In
every displayed frame, each full color pixel needs values
for all color components in the pixel. Therefore, the
input signal must be processed and sampled such that
there is a one-to-one correspondence between an input
sample and a pixel of the display. In the addressing
process, each input sample (RGB triplet for a specific
location in the image) is directed to a particular full
color pixel.
processing sampling addressing light
emission
Figure 2: Basic diagram of signal flow in a matrix display
system
This paper aims to determine a value for each subpixel
such that the quality of the displayed image is maximized.
In other words, the input image is reconstructed
in the best possible way. Even if there is no one-to-one
correspondence between the samples in the input image
and the (sub)pixels of the matrix display, which is
a common situation.
The subpixels represent a higher spatial resolution
(three times in horizontal direction for the arrangement
shown in Figure 1) than the (full color) pixel resolution
[1, 2, 3, 4, 8]. However, we cannot neglect the color
of the subpixels. Simply addressing the display with a
monochrome image of a higher resolution would result
in serious color artifacts, as will be shown in Section
3.1.
In references [2, 3, 8], it is suggested to take a weighted
average of pixels in the triple-resolution input image.
This will prevent color errors by spreading out the
intensity of each input pixel equally over red, green
1
Michiel A. Klompenhouwer Subpixel Image Scaling for Color Matrix Displays
and blue subpixels, creating ‘virtual’ or ‘logical’1 pixels.
These subpixel rendering methods profit from an
apparent resolution increase.
Given these subpixel rendering algorithms, the interesting
question arises how much extra resolution is actually
gained. This typically involves psychophysical
modelling of human vision [5], or subjective testing
[4, 12, 15]. Since this paper mainly deals with the
consequences of the subpixel structure from a signal
processing point of view, a detailed treatment of this
question is considered outside the scope of this paper.
In the remainder of this paper, we will first analyse, in
the frequency domain, the effect of the subpixel structure
in displays. This analysis shows that we the increase
in perceived resolution can be implemented at
low additional cost in a flexible image scaler, allowing a
straightforward control over the trade-off between color
errors and sharpness. Finally, we will show results, on
natural images and on text, and draw conclusions.
2 The resolution of a color matrix display
Let us calculate the frequency spectrum to analyse an
image that is displayed on a matrix display with a vertical
stripe subpixel arrangment2. Clearly, a matrix
display can only approximate the light intensities corresponding
to the original (space-continuous) image,
since it generates a space-discrete signal.
Figure 3 shows the basic system model of the signal
flow in a vertical stripe matrix display. Let us assume
that the RGB signals are sampled at positions (x, y)
that correspond to the center of a full color pixel at
the display. These sampled signals, (rs, gs and bs),
form the input to the addressing process. This process
translates into a spatial offset (delay) for each color.
In this case −x/3 for R, and +x/3 for B, if x is
the horizontal full color pixel distance. The translation
of the signal amplitude to the physical light emission
intensity, i.e. the reconstruction process, is carried out
by the light emitting/transmitting area (aperture) of
each (sub)pixel.
Formally, the sampling process results from multiplying
the continuous signal with a 2-D series of -impulses,
at intervals of x and y (vertically) [16]:
[attachment=27784]
Abstract
The perceived resolution of matrix displays increases
when the relative position of the color subpixels is taken
into account. ‘Subpixel rendering’ algorithms are being
used to convert an input image to subpixel-corrected display
images. This paper deals with the consequences of
the subpixel structure, and the theoretical background
of the resolution gain. We will show that this theory
allows a low-cost implementation in an image scaler.
This leads to high flexibility, allowing different subpixel
arrangements and a simple control over the trade-off
between perceived resolution and color errors.
1 Introduction
To generate full color images, matrix displays like
Plasma Display Panels (PDPs) and Liquid Crystal Displays
(LCDs) use three spatially displaced primary
color subpixels (red, green and blue, RGB) per full
color pixel. These subpixels are arranged in some repeating
pattern, like the ’vertical stripe’ arrangement,
shown in Figure 1.
a b c
Figure 1: In a matrix display (a), each full color pixel
consists of several primary color subpixels (b), here in
the ‘vertical stripe’ arrangement. The question arises
whether these subpixels can give extra resolution when
the grouping into full color pixels is released, as illustrated
by this simple example ©.
Each subpixel is given an intensity corresponding to the
value of the color component at the corresponding full
color pixel location in the image. When the subpixels
michiel.klompenhouwer[at]philips.com
†g.de.haan[at]philips.com
are small enough, they are not individually visible at a
normal viewing distance, so that the viewer will only
perceive the resulting color (tristimulus value) for that
location in the image: so-called ‘color blending’ occurs.
Figure 2 shows the basic diagram of the signal flow in a
matrix display, as it will be discussed in this paper. In
every displayed frame, each full color pixel needs values
for all color components in the pixel. Therefore, the
input signal must be processed and sampled such that
there is a one-to-one correspondence between an input
sample and a pixel of the display. In the addressing
process, each input sample (RGB triplet for a specific
location in the image) is directed to a particular full
color pixel.
processing sampling addressing light
emission
Figure 2: Basic diagram of signal flow in a matrix display
system
This paper aims to determine a value for each subpixel
such that the quality of the displayed image is maximized.
In other words, the input image is reconstructed
in the best possible way. Even if there is no one-to-one
correspondence between the samples in the input image
and the (sub)pixels of the matrix display, which is
a common situation.
The subpixels represent a higher spatial resolution
(three times in horizontal direction for the arrangement
shown in Figure 1) than the (full color) pixel resolution
[1, 2, 3, 4, 8]. However, we cannot neglect the color
of the subpixels. Simply addressing the display with a
monochrome image of a higher resolution would result
in serious color artifacts, as will be shown in Section
3.1.
In references [2, 3, 8], it is suggested to take a weighted
average of pixels in the triple-resolution input image.
This will prevent color errors by spreading out the
intensity of each input pixel equally over red, green
1
Michiel A. Klompenhouwer Subpixel Image Scaling for Color Matrix Displays
and blue subpixels, creating ‘virtual’ or ‘logical’1 pixels.
These subpixel rendering methods profit from an
apparent resolution increase.
Given these subpixel rendering algorithms, the interesting
question arises how much extra resolution is actually
gained. This typically involves psychophysical
modelling of human vision [5], or subjective testing
[4, 12, 15]. Since this paper mainly deals with the
consequences of the subpixel structure from a signal
processing point of view, a detailed treatment of this
question is considered outside the scope of this paper.
In the remainder of this paper, we will first analyse, in
the frequency domain, the effect of the subpixel structure
in displays. This analysis shows that we the increase
in perceived resolution can be implemented at
low additional cost in a flexible image scaler, allowing a
straightforward control over the trade-off between color
errors and sharpness. Finally, we will show results, on
natural images and on text, and draw conclusions.
2 The resolution of a color matrix display
Let us calculate the frequency spectrum to analyse an
image that is displayed on a matrix display with a vertical
stripe subpixel arrangment2. Clearly, a matrix
display can only approximate the light intensities corresponding
to the original (space-continuous) image,
since it generates a space-discrete signal.
Figure 3 shows the basic system model of the signal
flow in a vertical stripe matrix display. Let us assume
that the RGB signals are sampled at positions (x, y)
that correspond to the center of a full color pixel at
the display. These sampled signals, (rs, gs and bs),
form the input to the addressing process. This process
translates into a spatial offset (delay) for each color.
In this case −x/3 for R, and +x/3 for B, if x is
the horizontal full color pixel distance. The translation
of the signal amplitude to the physical light emission
intensity, i.e. the reconstruction process, is carried out
by the light emitting/transmitting area (aperture) of
each (sub)pixel.
Formally, the sampling process results from multiplying
the continuous signal with a 2-D series of -impulses,
at intervals of x and y (vertically) [16]: