INTERACT FORUM
More => Old Versions => Media Center 17 => Topic started by: Matt on December 04, 2011, 11:04:02 pm
-
A while back, somebody (maybe SamuriHL) asked for more control over image scaling in Media Center.
I meant to respond at the time, but didn't.
The image engine in Media Center has a variety of image resizing methods depending on the quality / performance required.
In high-quality mode, image stretching is done with bilinear resampling and image shrinking is done with supersampling.
My understanding is that supersampling is superior to Lanczos, Bicubic, etc. for extreme shrinks since it doesn't skip source pixels when building a destination pixel. The trade-off is that it can be slower.
The image engine is used for all 2d graphics in the program -- all of Standard View, texture creation in Theater View, subtitles, etc.
When viewing actual images (like JPEG files) in Media Center, a combination of the GPU and CPU is used. When moving, zooming, etc. the GPU is used to resize and display the image. When the image stops moving, a high-quality pixel-perfect overlay fades in over the top rendered by the JRiver algorithms listed above.
-
Perhaps it was TheLion here?
http://yabb.jriver.com/interact/index.php?topic=66587.0 (http://yabb.jriver.com/interact/index.php?topic=66587.0) (search "scaling").
-
Yes indeed, it was me.
It would be nice to get a choice - eg. opt for Lanczos (3/4) instead of "supersampling".
I am not familiar with the term supersampling in that context. Supersampling generally means doing an upscale, processing within that increased resolution, and than downscale to output resolution.
In my case I have 22MP files from my Canon 5d Mark II. I need my viewer to do the best possible downsampling to 1080p (for viewing with front projection) and be color managed. Editing is done in Lightroom - MC with its database features would be a nice platform to do the viewing.
-
I think the naming is a little squishy. Here's Intel describing Supersampling (http://software.intel.com/sites/products/documentation/hpc/ipp/ippi/ippi_ch12/functn_SuperSampling.html).
And here you can see them say "popular algorithms from the simplest – nearest neighbor, bilinear – to the more sophisticated – supersampling (the best image quality for reducing image size without any artifacts), different cubic filters, and so-called Lanczos3 filter" (http://software.intel.com/en-us/articles/the-intel-avx-realization-of-lanczos-interpolation-in-intel-ipp-2d-resize-transform/).
We don't use Intel IPP for image sizing, but what they describe is quite similar to our approach.
Please feel free to compare the sizing to Photoshop or any other program. In my experience, JRiver compares favorably. But if you find cases you think could be improved, please post details.
Thanks.