Striping error for Roche/Ventana .bif files

[Derek Magee]

[Moving discussion to git issue tracker]

(Will try to find some time to take a look)

-----Original Message-----
From: openslide-users [mailto:openslide-users-bounces+d.r.magee=leeds.ac.uk at lists.andrew.cmu.edu] On Behalf Of Benjamin Gilbert
Sent: 24 May 2017 07:17
To: openslide-users at lists.andrew.cmu.edu
Subject: Re: Striping error for Roche/Ventana .bif files

On Mon, May 22, 2017 at 01:47:14PM +0000, Derek Magee wrote:
> I wrote the messages (below) a couple of months back and got no reply 
> from the list.

Hi Derek,

I'm pretty swamped at the moment and far behind on email.  It looks as though you are indeed seeing #132.  You may be able to convince yourself one way or the other by setting the OPENSLIDE_DEBUG environment variable to "tiles" and inspecting the image at different zoom levels.

> There is some discussion there about needing a global optimisation of 
> position.  Optimisation is sort of my thing, so maybe I can help here?

Help would be greatly appreciated.  I don't anticipate having time to work on this issue anytime soon.

> I'm assuming we're talking about an optimisation based on the tile 
> positions in the metadata, not on the image content.

Right.  The scanner software has already computed the pairwise alignments based on the image content.

> I suspect this could be implemented either with no additional 
> libraries or using something with a generic linear least squares 
> solver like CLAPACK (BSD licenced).

If we could use either a commonly-available open-source library or a small amount of new OpenSlide code, that would be ideal.

> Consider the following approach:
> 
> Set tile 0,0 to position 0,0 (or other arbitrary position) For each 
> tile in row 0 set position based on tile to it's left For each tile in 
> column 0 set position based on tile above Then loop through remaining 
> tiles in row then column order:
>     Set position based on average of tile above and left (plus 
> relevant offsets)

That might be better than the global average we use now, but it wouldn't really solve the problem.  The edge tiles are likely to have few good features, so their pairwise alignments will be approximate at best (with correspondingly low confidences).  This approach weights them heavily, at the expense of the inner tiles with higher confidences, where it's also relatively important that we get the alignment correct.

> P.S.  I only just discovered the GIT issue tracker.  Should this 
> discussion be there?

Yep, that'd probably be a better place for it.

Thanks,
--Benjamin Gilbert
_______________________________________________
openslide-users mailing list
openslide-users at lists.andrew.cmu.edu
https://lists.andrew.cmu.edu/mailman/listinfo/openslide-users