Custom vendor format

[Benjamin Gilbert]

Hi Tim,

On 05/12/2014 04:05 PM, Blattner, Timothy J. wrote:
> So I was wondering if it would be possible for me to add support into
> OpenSlide to visualize images that are constructed using a CSV file that
> contain ‘path_to_file, x_pixel_position, y_pixel_position’.

You should be able to create a vendor driver for such a format without 
too much difficulty.

     https://github.com/openslide/openslide/wiki/AddingASlideFormat

 From your description, it sounds as though the output of the scanner is 
a single zoom level with no downsampling.  With one small exception, 
OpenSlide never synthesizes downsampled levels, so the API would report 
a single-level slide.  Naturally, that may not be useful to you.

If downsampled tiles could be precalculated and stored during scanning, 
that would allow OpenSlide to present a multi-level slide suitable for 
zooming.  In that case, either you will end up with very small tiles (in 
highly-downsampled levels) or you will need to merge tiles during 
downsampling.  That, in turn, would require you to resolve overlaps 
during downsampling, which may be more work than you want to do.  (We do 
support a couple of formats which concatenate and downsample adjacent 
tiles *without* first resolving overlaps, and I wholeheartedly advise 
not doing that.)

(The small exception is that OpenSlide can efficiently downsample JPEG 
tiles by 1/2, 1/4, or 1/8 at runtime, as a concession for Hamamatsu VMS. 
  That approach uses less CPU than a full decode, but the same amount of 
disk I/O.)

Also note that the tilemap grid expects to receive both the position and 
dimensions for each tile, so you'll want the metadata to include tile 
dimensions.

Does that help?  Feel free to ask questions on this list if you run into 
trouble.

> Now the tricky part with this technique is we would need to do
> blending on the overlapping regions, so either linear, average, or
> overlay; however, I think that it will not be expensive to do
> blending on the fly.

We currently use CAIRO_OPERATOR_SATURATE, so in effect we're only using 
pixels from the frontmost (bottom-right) overlapping tile except at 
seams.  But, since vendor drivers draw directly to the Cairo context, 
you should be able to use any of the Cairo compositing operators.

--Benjamin Gilbert