Re: Rendering quality (subsampling)

Thanks for the comments Martin (and Jeremias).

I ended up going with the oversized rendering and scaling down the
resulting image.  Interestingly, that too had surprising, unwanted effects
- I was using Java 2d.  Even testing on Gimp, using high quality scaling,
the circle that overlaps four pixels evenly, when scaled down from 400 x
400 (or 100 x 100) pixels to 4 x 4 ends up with uneven intensities.  I
didn't investigate deeply but there must be some degenerate case with the
interpolation when you get down to just a few pixels.

So I wrote my own scaling software.  It considers my 400 x 400 image to be
made of 4 "superpixels" that are 100 x 100.  Then I get a single value for
each of my superpixel (taking average intensity) and get a correct 4 x 4.
 Basic supersampling.  Works well.

Steve

On Thu, Dec 22, 2011 at 4:42 AM, Martin J <[email protected]> wrote:

> FWIW, my experience is that regrettably, the HW/OS combo is key to the
> image rendering quality; I had a servlet app I developed on Mac OS X
> that produced SVG images on demand that had to be returned to the Web
> client as PNG. On the Mac, the rendering was extremely good, but when
> run on a PC/Linux server, it produced unacceptable images. In our
> case, this issue was mostly that of poor text rendering at small type
> sizes, due to bad anti-aliasing. After a lot of testing, we decided to
> get a Mac dedicated to image rendering!   YMMV.
> best
> Martin
>
> On 14 December 2011 16:01, Steve Maher <[email protected]> wrote:
> > Hi,
> >
> > I'm looking for a high-quality SVG renderer.  Batik's rasterizer seems
> to be
> > the best out of the three I've tried, but I need it to do better.  I've
> > tried all the image-rendering, shape-rendering, color-rendering hints
> > without much (any?) change.  I've searched the source - mainly to try to
> > increase any subsampling quantities - but had no luck.
> >
> > I'm dealing with small circles that need to be rendered "correctly" for a
> > custom projector in our lab at work.
> >
> > For example, for a circle of radius .5 pixels (no stroke), and centered
> in
> > the middle of a pixel (in other words, a circle touching the edges of a
> > pixel), the rendered pixel intensity should be roughly PI/4 times the
> > intensity of the circle (because the ratio of circle area to square area
> is
> > PI/4).
> >
> > Also, if I rotate the circle by 45 degrees (rotation centered on corner
> of
> > pixel), there will be a 4 pixel square used to render the circle, with
> most
> > intensity split between two adjacent pixels and much less intensity split
> > between the other two adjacent pixels.
> >
> > Batik comes close, but the intensity is high in the first case and in the
> > second case, the intensity is not symmetric between the adjacent pixels
> > (like the rotation wasn't exactly 45 degrees).
> >
> > One option I am exploring is to artificially enlarge the scene, render
> by 1
> > pixel circle across, oh, 100 pixels, and then shrink the image.
>  Effectively
> > increasing the subsampling of the circle to determine pixel intensity.  I
> > was hoping to be able to do that natively in the Batik source - or use
> some
> > other SVG renderer.
> >
> > Any tips are appreciated -
> >
> > Regards,
> > Steve
> >
> >
>
>
>
> --
> From my MacBook Pro
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected]
> For additional commands, e-mail: [email protected]
>
>