connecting scanned canvas paintings?

[David Melik]

I scanned a painting of mine (that I think is on canvas that is over cardboard) in scans of four all overlapping sections, and followed the tutorial on connecting scanned pieces. It works almost ok if I do it automatically (without the tutorial), but Hugin seemed to make bands/waves of significantly differing colour tone that are not on the original, but are sort of parallel--but curved--in relation to the painting's horizon. If I follow the tutorial, then one side of the image gets connected maybe ok, but the other side has some 3D warp that seems to have little to do with the original. I am not an expert on Hugin or anything similar, but what could be happening or could I be doing wrong for this case?

The painting itself can be called flat, but the medium was woven, but not sanded several times after priming to make it flat, so the weave still shows up a bit when looking extremely closely. That seems to make it hard for Hugin to detect all the points (since the weave casts shadows) though on one or two sets it detects tens of points... maybe the problem is it is in four sections, more than the points... well, when I add points in the ones that were not connected, Hugin cannot automatically detect them, but as I said, it did connect two sets of the four with many automatic points... it is strange. Anyway, I am not really sure what went wrong or what to do. I can upload what I am trying to connect somewhere, such as on my own server (which would be slow to download the about 1 GB, but I could maybe use  deviantART) if that would help... if anyone is willing to help.

I hope this is not too complicated, but I hope that if no one had done something like this, someone involved with Hugin could figure out how. I know it is possible with this sort of software, because I had the same painting professionally scanned & connected as a poster for me many years ago, and the graphician did it perfectly.

[Terry Duell]

Hello David,

Some comments below.

This might sound like a silly question, but have you carefully checked all
the control points that Hugin has found?
It is possible that the weave (which I assume is pretty constant) is
causing the control point detector to make errors.
The process described in the tutorial should work OK, if the control
points are properly placed.
I would be happy to have look at the project but at full scale it is too
big for me to contemplate a download.
Could you downsize each of the images in the set, to (say) something like
800x800, which would make the download a possibility.
The only issue I foresee with downsized images is loss of detail in the
weave, but that is just a guess.

Cheers,
--
Regards,
Terry Duell

[David Melik]

On Saturday, 5 October 2013 00:00:47 UTC-7, Tduell wrote:

[...] This might sound like a silly question, but have you carefully checked all  

the control points that Hugin has found?

No; it seems to connect mostly/all ok but just projects half the picture into 3D space rather than leaving it 2D. I will see if I can check soon anyway.

 

It is possible that the weave (which I assume is pretty constant) is  
causing the control point detector to make errors.
The process described in the tutorial should work OK, if the control  
points are properly placed.
I would be happy to have look at the project but at full scale it is too  
big for me to contemplate a download.

Could you downsize each of the images in the set, to (say) something like  
800x800, which would make the download a possibility.
The only issue I foresee with downsized images is loss of detail in the  

weave, but that is just a guess.[...]

I made a mistake; it is 448 MB, or about 238 in greyscale. I could also send it on a CD. I need the picture as large as possible, and my camera already takes larger photos than it would be resized, but if it would help to debug and/or improve Hugin so people could connect these sorts of pictures, I would like to do whatever I can.

David

[Terry Duell]

On Wed, 09 Oct 2013 09:10:26 +1100, David Melik <dchm...@gmail.com> wrote:

> On Saturday, 5 October 2013 00:00:47 UTC-7, Tduell wrote:
>

[snip]

>> Could you downsize each of the images in the set, to (say) something
>> like
>> 800x800, which would make the download a possibility.
>> The only issue I foresee with downsized images is loss of detail in the
>> weave, but that is just a guess.[...]
>>
>
> I made a mistake; it is 448 MB, or about 238 in greyscale. I could also
> send it on a CD. I need the picture as large as possible, and my camera
> already takes larger photos than it would be resized, but if it would
> help
> to debug and/or improve Hugin so people could connect these sorts of
> pictures, I would like to do whatever I can.

448 MB is still too much for me. Something closer to 5 MB for the set
would be good.
My suggestion of making downsized images available was to test the
process, which you could then try using your full sized images. I have
worked with downsized images a few times and the result has always been
almost identical.

[David Melik]

On Tuesday, 8 October 2013 15:48:14 UTC-7, Tduell wrote:

448 MB is still too much for me. Something closer to 5 MB for the set  
would be good.
My suggestion of making downsized images available was to test the  
process, which you could then try using your full sized images. I have  
worked with downsized images a few times and the result has always been  
almost identical.

Ok; I shrunk them to about 800x1134 (they are not squares), and the weave is still visible, and they are 5.9 MB total... I could make the larger dimension 800 if you like, or cut off part, or if this is small enough, I could post them somewhere or send them. At this size, I would rather send; I do not usually make scans this size public, but if I made them smaller, I could post it.

David

[Terry Duell]

Hello David,

OK, I understand you don't want to make them public, but you could put
them on Dropbox, or similar, and then send me the URL off list. Probably
easier than sending them via snail mail, and you can remove them from
dropbox immediately I let you know that I have them OK. Whatever suits you.
If you want to send, contact me off list and I'll give you my details.

[paul womack]

Terry Duell wrote:

>
> OK, I understand you don't want to make them public, but you could put them on Dropbox, or similar, and then send me the URL off list. Probably easier than sending them via snail mail, and you can remove them from dropbox immediately I let you know that I have them OK. Whatever suits you.
> If you want to send, contact me off list and I'll give you my details.

The low-res could also have some "damage" added; a quick scribble in Photoshop or Gimp would serve.

BugBear

[David Melik]

On Thursday, 10 October 2013 20:29:23 UTC-7, Tduell wrote:

OK, I understand you don't want to make them public, but you could put  
them on Dropbox, or similar, and then send me the URL off list. Probably  
easier than sending them via snail mail, and you can remove them from  
dropbox immediately I let you know that I have them OK. Whatever suits you.
If you want to send, contact me off list and I'll give you my details.

I never really used anything like Dropbox, which would all be harder for me than email (and even putting them on my own FTP or web server, and I did not want to put them on the latter), though Google Drive is almost as easy. You can add me on Google+ or write dchm...@hipplanet.com, and if with a Gmail address, let me know if you prefer Google Drive or email.

D

[David Melik]

OK, I understand you don't want to make them public, but you could put  
them on Dropbox, or similar, and then send me the URL off list. Probably  
easier than sending them via snail mail, and you can remove them from  
dropbox immediately I let you know that I have them OK. Whatever suits you.
If you want to send, contact me off list and I'll give you my details.

Tduell,

Thanks for letting me mail you the scans and showing me it could work. After your description, I rescanned better, but did not get similar results: http://mathematicon/david/art/scan_problem.jpg. I think you said I could email you again, but I do not want to send too many emails, so am posting... just wondering what I did wrong this time. Actually, after rescanning in a better way (though I should probably redo it even again, to be more precise), Hugin did not automatically connect them; I had to go by the tutorial on connecting scanned images, including adding the points.

D

[Terry Duell]

Hello David,

On Wed, 16 Oct 2013 17:48:22 +1100, David Melik <dchm...@gmail.com> wrote:


>
> Thanks for letting me mail you the scans and showing me it could work.
> After your description, I rescanned better, but did not get similar
> results: http://mathematicon/david/art/scan_problem.jpg. I think you
> said I
> could email you again, but I do not want to send too many emails, so am
> posting... just wondering what I did wrong this time. Actually, after
> rescanning in a better way (though I should probably redo it even again,
> to be more precise), Hugin did not automatically connect them; I had to
> go by the tutorial on connecting scanned images, including adding the
> points.
>

By all means contact me again, I'm happy to try to help.
I didn't have any success with the link, it always ended up at
<mathematicon.com>.
One approach you might try, if you would like to, is to pre-align your
images prior to running CPFind.
Load your images, and note the numbering that hugin gives them.
Open the preview window, and select the "move/drag" tab.
You are now able to choose which images are displayed using the "displayed
images" area. Select only one image and drag it into a position that
roughly approximates where it should be in the overall layout. Use the
cross hairs as a guide for position and overlap if you have a two by two
arrangement.
Add another image, dragging it into position, with approx overlap. Do that
for all images.
In the editor window, choose "Hugin's CPFind (prealigned)", and create
control points.
If that doesn't help automatically find suitable control points then you
may have to resort to manually setting your control points, although from
my knowledge of your initial set of images I wouldn't expect an issue if
you have sufficient overlap.

See how you go, and if you would still like more help feel free to email
me an archive of downsized images from your new scan.

Hope that helps.

[David Melik]

On Wednesday, 16 October 2013 01:37:31 UTC-7, Tduell wrote:

By all means contact me again, I'm happy to try to help.
I didn't have any success with the link, it always ended up at  
<mathematicon.com>.

Oops, it should have been http://www.mathematicon.org/david/art/scan_problem.jpg

I just did a lot of other graphics editing, and now it is pretty late, so I will reread and hopefully try some of what you said tomorrow or soon.

Thanks again;
Best Wishes,
David

 

[Terry Duell]

Hello David,

On Wed, 16 Oct 2013 22:34:57 +1100, David Melik <dchm...@gmail.com> wrote:

>
>
> On Wednesday, 16 October 2013 01:37:31 UTC-7, Tduell wrote:
>>
>> By all means contact me again, I'm happy to try to help.
>> I didn't have any success with the link, it always ended up at
>> <mathematicon.com>.
>>
>
> Oops, it should have been
> http://www.mathematicon.org/david/art/scan_problem.jpg
>

OK, I can now see your stitched image.
Not sure why you have the band of vertical lines, maybe someone else has
seen this before.
I assume this is a stitch of the original scanned images, not downsized
images.
A stitch of scanned images should look nice and rectangular.
It also seems odd to me that the stitch is distorted and the lines are
straight and vertical. I wonder if the lines are being introduced by the
blender. Which blender did you use, enblend or multiblend, and which
version?

[David Melik]

On Wednesday, 16 October 2013 15:16:17 UTC-7, Tduell wrote:

OK, I can now see your stitched image.
Not sure why you have the band of vertical lines, maybe someone else has  
seen this before.
I assume this is a stitch of the original scanned images, not downsized  
images.
A stitch of scanned images should look nice and rectangular.
It also seems odd to me that the stitch is distorted and the lines are  
straight and vertical. I wonder if the lines are being introduced by the  
blender. Which blender did you use, enblend or multiblend, and which  
version?

Maybe the extra vertical lines happened if the stitching had not finished, but I mainly wanted to show it was not rectangular. Of course I did use the original scanned images, but as the original picture is about 458x610 mm (18"x24"), I of course shrunk the posted one. I did get multiblend working, but this time I used enblend, as suggested. It was late at night when I did this, and though I think I followed the tutorial, I am not 100% sure I chose 'mosaic' in the optimizer, or if I did, it maybe said some complicated message and had to change settings that were set before that stage (setting most the r,X,Y,Z to zero), so I may not have used 'mosaic'. I can try this again earlier in the day and pay closer attention.