Pixel Borrowing, ClearType and Antialiasing
During his keynote address at the most recent COMDEX conference,
Microsoft President Bill Gates announced "a breakthrough technology",
called ClearType, to display type on LCD screens. The story was covered
thoroughly in the mainstream media as well as by the technical press and
the omnipresent Microsoft press relations machinery.
This article is partially about ClearType. But mostly it is about the problem
ClearType is intended to address: the low visual quality of most of the
information presented on computer screens. Here is a summary of what
I think (the same topics are also discussed in a recent article in the New
As this outline should make clear, I am no friend of ClearType. I agree with
- The text and vector graphics in our screens are of very poor quality,
i.e., the problem that ClearType addresses is all too real.
- ClearType appears not to be either new or a breakthrough. At best
it produces some improvement in very particular circumstances.
- Quality improvements require a substantial increase in display
resolution and the use of antialiasing, but increasing the resolution of
displays breaks all our software.
- Display resolution will increase in the future, but it would happen
faster if Microsoft pushed for it now. ClearType may well be a
distraction from a real solution.
the assertion made by Bill Gates in the ClearType announcement that the
type we read on our screens today is in need of immediate and substantial
improvement - although I would immediately add that vector graphics are
also equally wretched and in need of similar improvement. I differ in my
prescription to deal with the problem: my bet is a combination of antialiasing
and higher display resolutions. I consider ClearType as a distraction. At best
it is a benign placebo for a serious problem that requires surgery. At worst it
my be an excuse to have us talk about Microsoft in a context unrelated to
ClearType or Pixel Borrowing?
The issue of ClearType's novelty - more precisely, the lack thereof - is
addressed by two online articles one in The New York Times and another
by Steve Gibson. Both point out that similar, if not identical, techniques
have been known for many years. The basic idea was used in the Apple II
computer, where it was known as "pixel borrowing" and was the subject
of a patent granted to Steve Wozniak, the Apple Computer co-founder.
More important than the question of novelty is whether pixel borrowing can
indeed improve substantially the visual quality of what we see in our screens.
The answer seems to be negative. In addition to some static examples of
pixel borrowing at work, Steve Gibson's web site also offers a small program
that allows you to experiment with the technique by using it on text set with
various fonts and your choice os sizes while varying certain rendering
parameters. Because there seems to be no available samples or demos from
Microsoft, the discussion below is based on Gibson's demo and my own
The limitations of Pixel Borrowing
Steve Gibson describes the limitations of pixel borrowing as follows:Another way to state these limitations is as follows: pixel borrowing will not
It only works on color LCD displays and does not work on high
resolution CRT displays.
It works well only for black shapes over a white background or
white on black - high color contrast in all three primary colors is
required for pixel borrowing to be effective.
It improves shapes in only one direction - the horizontal direction
for most LCD displays.
help for colored text or black text over a colored background, will not help
for most graphics, and will help very little for Kanji and other complex
scripts that, unlike roman text, are helped little by resolution increases in
only one direction.
In addition to these limitations, I am a lot less enthusiastic than Microsoft
about the basic promise of pixel borrowing even in the case where it does
help, that of black latin text over a white background. My experiments and
Steve Gibson's demo program show that the purest forms of pixel borrowing
exhibit substantial color artifacts, similar to the jaggies of typical display text
but at the color subpixel level. Careful (and computationally expensive)
processing can reduce or eliminate these color artifacts but at the expense of
spreading the intensity of each color subpixel across its neighbors, reducing
the sharpness of edges which Microsoft claims as the main advantage of
ClearType over antialiasing.
Everything I have seen and heard tells me that the 200-300% increase in
horizontal resolution claimed by Microsoft's literature for ClearType is based
on little more that hopeful arithmetic. The argument seems to run along the
following lines. There are three color subpixels per pixel and if they were fully
utilized the resolution would increase by 300%; but one primary is likely to be
less helpful than the others (blue: we are less sensitive to it than to red or
green) and some of that resolution needs to be "spent" to reduce the color
artifacts, so maybe 200% is a more reasonable expectation. Using my own
arithmetic to quantify the result of pixel borrowing is unlikely to shed much light
on the matter, so I will refrain from it. Whatever the numerical value, I would not
use "breakthrough" to describe the improvement produced by pixel borrowing.
In the ClearType hoopla, antialiasing has played
the part of the villain. This is
strange because the techniques needed to reduce color artifacts when using
pixel borrowing are basically antialiasing techniques. Also a lot of people have
to go considerably out of their way to use antialiased graphics and yet do it
regularly. Most 2D web graphics (e.g., the Ductus logo at the top of this page)
start their lives as vector graphics. While being created with a package like
Adobe Illustrator or Macromedia Freehand, these 2D graphics are shown on
screen with the usual aliasing artifacts: jaggy lines, inaccurate text spacing,
inaccurate line thickness, disappearing parts, etc. Once the artwork is finished
and before it is used in a web page, the graphic is turned into an antialiased GIF
or JPEG file to eliminate these artifacts often with dramatic improvements as a
result. So why is antialiasing the villain of the story?
The principal argument against antialiasing is
related to text.
rendered antialiased text looks somewhat fuzzy at reading sizes when
viewed at typical reading distances (around 18 inches) on typical displays.
The fuzziness is particularly noticeable on LCD displays, the one type of display
where pixel borrowing helps. One could easily weight the fuzziness of antialiased
text against the many ills of the aliased text we read today - jaggies, poor spacing,
distorted weight, mangled character shapes, etc. - and decide that antialiasing is
a winner in the trade. On a good CRT display there is little doubt that antialiasing,
used correctly, leads to the best results for both text and vector graphics. Even
on a good LCD screen I find the issue of which end of the bargain is preferable
- jaggies or fuzzies - debatable. There is no question, however, that we would be
better off without jaggies and without fuzziness.
Getting rid of both the fuzziness and the
aliasing is not hard: evidence from vision
research suggests that when a pixel subtends an arc of less than 1.4 minutes in our
visual field, most of the fuzziness produced by antialiasing is not visible. To put that
fact in more concrete and familiar terms screens of around 140 dpi will do the trick
assuming a reading distance of 18 inches. This is approximately 50% increase in
resolution from typical 90 dpi screens of today, a large increase without a doubt,
but well within what is feasible today. Note that the use of antialiasing is a necessary
ingredient in this prescription: aliasing artifacts are clearly visible at 140 dpi -
just think back to the jaggies and uneven stem weights of roman type in 300 dpi
laser printer output. It appears that in this story, the bad guy is destined to win.
Could This Really Work?
That all our quality problems could be solved
by increasing current display
resolutions by 50% in each direction and using antialiasing properly sounds almost
too easy to be true, so let me anticipate a few objections and explain why they do
not invalidate my argument:
So the pieces are falling neatly into place except for one. Our problem is software -
all of it: the GUIs of our operating systems, our applications, all these raster images
used in our web pages. Eventually this problem will also be solved and we will be able
to use displays with resolutions above 150 dpi for our everyday work. There is no
question, however, that this would happen much faster if Microsoft got in front of the
parade. Unfortunately, it seems like we will get ClearType instead.
What Bill Gates Should Have Said at COMDEX
I am quite glad that Bill Gates decided to use
the COMDEX keynote address to point
out that what we see on our screens today is of a deplorable level of quality. I would
have addressed the same issue, had I been given the chance. His proposal of ClearType
as the solution was off the mark, however. Instead of the ClearType "breakthrough" he
should have announced that Windows 2000 would use a GUI that is fully resolution
independent - suitable for displays of resolutions of 150 dpi and above - and that future
versions of all the Microsoft applications would work with the resolution independent
These are files produced with Photoshop and Illustrator
to simulate pixel borrowing.
Although Illustrator and Photoshop are very nice tools for the sort of work for which
they designed, using them for this type of subpixels hacking is a little like sculpting with
a jackhammer. No claim is made that this is the best the pixel borrowing technique
can produce, hence the title of sampler bestowed on the collection. Still, I think that
these examples illustrate far better than words the potential and pitfalls of the
What kind of LCD screen do I have?The actual pixel borrowing examples:
How the examples were producedNearly vertical lines: the best situation for pixel borrowing
Vertical lines: color artifacts galore
A simple vector graphics example
A very small text sample
A more complex vector illustration
Naiman, Avi.C and Makous, Walter, "Undetected gray strips displace perceived edges
non linearly". J. Opt. Soc. Am. A/ Vol. 10, No. 5/ may 1993.
Copyright © 1997-2002 Ductus Inc.