lamaster@pioneer.UUCP (06/08/87)
I have heard that it is possible to represent lines of less than 1 pixel in
width with greyscale or color images using data filtered from a higher
resolution image or object. I have never seen either the algorithms or the
theoretical basis for an algorithm. [An example would be a representation,
using a greyscale picture, of a line less than one pixel or dot width in size].
I have looked in standard graphics textbooks and find no mention of this [but
I could have missed something or not understood what I was looking at]. Does
anyone know of any references for this? Are there any devices out there which
make use of it [e.g. a monitor with a much higher resolution frame buffer than
the tube and a filter to reduce the resolution before output to the tube]?
Hugh LaMaster, m/s 233-9, UUCP {seismo,topaz,lll-crg,ucbvax}!
NASA Ames Research Center ames!pioneer!lamaster
Moffett Field, CA 94035 ARPA lamaster@ames-pioneer.arpa
Phone: (415)694-6117 ARPA lamaster@pioneer.arc.nasa.gov
"In order to promise genuine progress, the acronym RISC should stand
for REGULAR (not reduced) instruction set computer." - Wirth
("Any opinions expressed herein are solely the responsibility of the
author and do not represent the opinions of NASA or the U.S. Government")bobr@zeus.UUCP (06/09/87)
If you've seen anything in those textbooks about anti-aliasing, then you've seen the methods used to perform this miracle of modern technology. Aliasing is the effect of producing false images by undersampling source images with higher spacial frequency than the display. A line of less than one pixel width exhibits this higher than displayable spacial frequency. The techniques of anti-aliasing sample the image at a high enough frequency to capture such details and then averages those samples in such a way that some representation remains when actually displayed. The general effect is one of blurring the details. Thus the less-than-a-pixel-wide line becomes a softened image that may still look like a line. -- Robert Reed, Tektronix CAE Systems Division, bobr@zeus.TEK
kens@hpldola.UUCP (06/10/87)
Years back, Jim Blinn had a head and shoulders shot of a woman which appeared to be off of a standard monitor, broadcast television. His comment: "You'd swear that you can see this..." while tracing along a single hair, floating free from the woman's head. The picture was only at 256x256 sort of resolution with greyscale, so it *couldn't* really be there. There's a lot that we don't know... Ken Shrum HP - Logic Design Operation hpldola!kens
nelson@esunix.UUCP (06/10/87)
in article <1713@ames.UUCP>, lamaster@pioneer.arpa (Hugh LaMaster) says: > > I have heard that it is possible to represent lines of less than 1 pixel in > width with greyscale or color images using data filtered from a higher > resolution image or object. I have never seen either the algorithms or the > theoretical basis for an algorithm. [An example would be a representation, > using a greyscale picture, of a line less than one pixel or dot width in size]. > I have looked in standard graphics textbooks and find no mention of this [but > I could have missed something or not understood what I was looking at]. Does > anyone know of any references for this? Are there any devices out there which > make use of it [e.g. a monitor with a much higher resolution frame buffer than > the tube and a filter to reduce the resolution before output to the tube]? > > Hugh LaMaster, m/s 233-9, UUCP {seismo,topaz,lll-crg,ucbvax}! > NASA Ames Research Center ames!pioneer!lamaster > Moffett Field, CA 94035 ARPA lamaster@ames-pioneer.arpa Disclaimer: Please don't take this as a sales pitch, it is intended for information only. There is one line drawing system currently on the market with a much higher effective resolution than the actual screen resolution. This is the Evans & Sutherland PS 390. The visible portion of the frame buffer is 1024 x 864 pixels with the line filtering occurring when the lines are drawn into the frame buffer. The effective resolution is 8192 x 6912. I and several other engineers here have done extensive research on anti-aliasing techniques and we know of no literature (available outside of our company :-> ) on this subject. Most of what I have seen on anti-aliasing lines is actually misleading. The usual methods in the top computer graphics books have roping on the lines which causes moire patterns when lots of lines are together and some temporal aliasing effects when lines are moved slowly. I have personally seen lines moving by only 1/8 of a pixel on a PS 390 where the motion could be seen with the unaided eye. On this machine, the higher effective resolution applies to line positioning but not to line width. To get lines with an actual effective line width of less than one pixel you will have to buy a CT-6. :-) ----- Scott R. Nelson Evans & Sutherland Computer Corporation UUCP Address: {decvax,ucbvax,ihnp4,allegra}!decwrl!esunix!nelson Alternates: {ihnp4,seismo}!utah-cs!utah-gr!uplherc!esunix!nelson seismo!usna!esunix!nelson "Smooth lines, they're not just for vector displays anymore."
turk@apple.UUCP (Ken "Turk" Turkowski) (06/12/87)
In article <1713@ames.UUCP> lamaster@pioneer.arc.nasa.gov (Hugh LaMaster) writes: >I have heard that it is possible to represent lines of less than 1 pixel in >width with greyscale or color images using data filtered from a higher >resolution image or object. I have never seen either the algorithms or the >theoretical basis for an algorithm. [An example would be a representation, >using a greyscale picture, of a line less than one pixel or dot width in size]. >I have looked in standard graphics textbooks and find no mention of this [but >I could have missed something or not understood what I was looking at]. Does >anyone know of any references for this? Are there any devices out there which >make use of it [e.g. a monitor with a much higher resolution frame buffer than >the tube and a filter to reduce the resolution before output to the tube]? The theory is really quite simple. Imagine an infinite resolution image over which a low-pass filter is passed, with cutoff frequency consistent with the Sampling Theorem. Sample the sucker, and what do you get? A line of one pixel in width whose amplitude is attenuated linearly as a function of the line width. Note that this model also accommodates sub-pixel positioning. Unfortunately, such lines do not look quite as good as 1 pixel wide lines because there are less quantized intensity values available to interpolate with. If the line gets real thin, you end up with a Bresenham. To do the computations in image space one could certainly use a Bresenham line-drawer at a higher resolution and then decimate (down-sample) the image, but I am am a proponent of object space algorithms. For references, theory, and algorithms, see: Turkowski, Keneth Anti-Aliasing through the use of Coordinate Transformations ACM Transactions on Graphics, Vol. 1, No. 3, July 1982 pp. 215-234. For thinner lines, you would just scale the line spread function. Another reference is: Gupta, S and Sproull, R Filtering Edges for Gray-scale Displays SIGGRAPH 1981 proceedings, vol. 15, no. 3, pp. 1-5. -- Ken Turkowski @ Apple Computer, Inc., Cupertino, CA UUCP: {mtxinu,sun,nsc,voder}!apple!turk CSNET: turk@Apple.CSNET ARPA: turk%Apple@csnet-relay.ARPA
jbm@aurora.UUCP (Jeffrey Mulligan) (06/12/87)
in article <347@esunix.UUCP>, nelson@esunix.UUCP (Scott Nelson) says: > > in article <1713@ames.UUCP>, lamaster@pioneer.arpa (Hugh LaMaster) says: >> >> I have heard that it is possible to represent lines of less than 1 pixel in >> width with greyscale or color images using data filtered from a higher >> resolution image or object. > > > Most of what I have > seen on anti-aliasing lines is actually misleading. The usual > methods in the top computer graphics books have roping on the lines > which causes moire patterns when lots of lines are together and some > temporal aliasing effects when lines are moved slowly. > These types of effects will occur when the color lookup table is not set up correctly to produce linear intensity values on the display (gamma correction). > > I have personally seen lines moving by only 1/8 of a pixel on a > PS 390 where the motion could be seen with the unaided eye. On > this machine, the higher effective resolution applies to line > positioning but not to line width. > On an 8-bit per pixel display, you should be able to move a line or point by 1/500 of a pixel. (Calculate the displacement of the centroid of the light distribution for these two pixel pairs: (x1,y1,i1)=(0,0,255) (x2,y2,i2)=(1,0,254) and (x1,y1,i1)=(0,0,254) (x2,y2,i2)=(1,0,255) At a viewing distance of 3-4 feet, a pixel will typically subtend about 1 minute of arc of visual angle. The smallest vernier offsets and/or motions that can be seen by the human eye are in the range of 3-5 arc seconds; certainly never less than 1 sec. (references upon request). To fully appreciate how small these numbers are, consider that the foveal cones (light collecting receptor cells) have a diameter of 30 src seconds. The point here is that, depending upon viewing distance, 8 bits of gray levels may not be necessary for effective anti-aliasing. It is interesting that anti-aliasing techiniques must necessarily INCREASE the width of lines. To represent lines having a width of less than one pixel, all you can do is reduce the contrast. Try this: with a one-pixel-wide line on the screen, increase your viewing distance. The line will appear thinner until its apparent width is equal to the optical point spread function of the eye. Increasing the viewing distance further will reduce the contrast of the line but not its width. So just make sure that you use a viewing distance such that the pixel size is less than or equal to 30 minutes of arc (approx. point spread diameter). > To get lines with an actual > effective line width of less than one pixel you will have to buy > a CT-6. :-) > I didn't get this joke :-( -- Jeff Mulligan (jbm@ames-aurora.arpa) NASA/Ames Research Ctr., Mail Stop 239-3, Moffet Field CA, 94035 (415) 694-5150
ferwerda@batcomputer.UUCP (06/16/87)
In article <1713@ames.UUCP> lamaster@pioneer.arc.nasa.gov (Hugh LaMaster) writes: >I have heard that it is possible to represent lines of less than 1 pixel in >width with greyscale or color images using data filtered from a higher >resolution image or object. I have never seen either the algorithms or the >theoretical basis for an algorithm. [An example would be a representation, >using a greyscale picture, of a line less than one pixel or dot width in size]. >I have looked in standard graphics textbooks and find no mention of this [but >I could have missed something or not understood what I was looking at]. Does >anyone know of any references for this? Are there any devices out there which >make use of it [e.g. a monitor with a much higher resolution frame buffer than >the tube and a filter to reduce the resolution before output to the tube]? > > I'm glad you asked. Two classic papers on this topic are Crow, F.C. The Use of Grayscale for Improved Raster Display of Vectors and Characters. Proc. SIGGRAPH 1978, 1-5. and Kajiya, J. and Ullner, M. Filtering High Quality Text for Display on Raster Scan Devices. Proc. SIGGRAPH 1981, 7-15. there's also a paper by Leler, W.J. Human Vision Anti-Aliasing and the Cheap 4000 Line Display. Proc. SIGGRAPH 1980, 303-318. for an exhaustive review of this whole topic, with more theory than you'd ever want, and with real empirical tests of its usefulness in static and dynamic raster displays, you can also read my thesis A Psychophysical Approach to the Aliasing Problem in Realistic Image Synthesis. Cornell University, 1987. by (me) James A. Ferwerda Happy Reading. Jim Ferwerda jaf@squid.tn.cornell.edu{.UUCP}