how does openslide do image downsampling

[石川拓海]

Hi, Benjamin and john,


Thank you for your help.


- Takumi Ishikawa
________________________________________

On 01/26/2015 10:02 PM, 石川　拓海 wrote:
> I'm using a NDPI format (Hamamatsu format) image. Does OpenSlide creates
> extra slide levels for this format?

For each real level in a Hamamatsu VMS or NDPI slide, OpenSlide tries to
create synthetic levels which are downsampled by factors of 2, 4, and 8
from the real level.  If a synthetic level would have the same
downsample factor as a real level in the slide file, the synthetic level
is skipped.

VMS slides have only two real levels, so synthetic levels will always be
created.  Some NDPI slides (older ones?) space the downsamples of their
real levels by a factor of 4, so every other level (plus the smallest
ones) will be synthetic.  Other NDPI slides space their real levels by a
factor of 2, so the only synthetic levels will be the last three.

> If so, do you know how does libjpeg do image downsampling? Gaussian
> pyramid?

As I understand it, libjpeg only partially decodes each JPEG MCU.  I
don't know the details.

--Benjamin Gilbert

________________________________________

On 27 January 2015 at 06:03, Benjamin Gilbert <bgilbert at cs.cmu.edu> wrote:
>> If so, do you know how does libjpeg do image downsampling? Gaussian
>> pyramid?

For 8x shrink, libjpeg does a simple 8x8 box filter in Y. I don't know
what it does for x4 and x2, but I would expect they would also be box
filters.

John