Around 19 o'clock on Sep 20, Allen Akin wrote:
> What I can say today is that using hardware full-scene antialiasing > techniques for rendering small text as geometry works surprisingly well on > a CRT. (4X supersampling with Gaussian filtering seems best at the moment. > I'd like to try the new FSAA techniques available on the Radeon 9700.) Digital typography started with that particular idea about 20 years ago. The problem is that CRT (and LCD) spot sizes are just too large to ignore when drawing text, so careful hinting and adjustments remain necessary to present text which is readable over long periods of time. Current theories state that when screens used in a desktop environment reach 200dpi, the spot size should be small enough for us to ignore it. Given the quadratic nature of screen sizes, I don't anticipate this kind of resolution in the next few years. > I also plan to try multipass rendering to implement high-quality > antialiasing filters with support greater than one pixel in area. The initial ClearType papers demonstrate this kind of filtering, and the theory is certainly compelling -- you should be able to provide near-nyquist frequencies on a CRT by doing full-screen filtering of this kind. However, we run (again) into the low resolution of the screen interacting with the human visual system. Using the (near) gaussian nature of the CRT spot, or the rectangular shape of the LCD element, we can produce edges with frequencies significantly higher than the nyquist limit. As the human visual system is tuned for these sharp edges, the glyph-shape deformation necessary to align glyphs to these edges must be balanced with the improvements in visual discrimination. The essential thing to remember is that there is no low-pass filter trimming the data after it exits the monitor; what would be 'aliases' in standard signal processing parlance turns into a feature in this environment, but a feature which requires significant pre-processing of the original object geometry to fully appreciate. Hence, font hinting is central to text presentation on the screen. TrueType provides mechanisms for the font designer to make the tradeoffs manually, Type1 provides some knobs for the designer to set and FreeType is slowly improving an automated system for fonts without hints. > I speculate that using geometry rather than pixmaps of antialiased > characters is a better approach in the long run, because with geometry > it's easier to apply small transformations, e.g. subpixel positioning or > scaling to preserve sharp vertical elements. Yes, using a non-geometric description of the glyphs makes any kind of transformation hopeless. While Render appears to favor the bitmap approach, the extension is intended only to provide a cache for images generated from geometry on the client side of the wire. Using bitmaps allows us to experiment with new rendering approaches without impacting the server or wire protocols. Keith Packard XFree86 Core Team HP Cambridge Research Lab _______________________________________________ Render mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/render
