Auto vs Manual for "Split Pages"

[JonEP]

Hi,

I'm trying out the new pre-release of ST. It's still great (no surprise!) -- I don't notice the difference too much, but I'll take your word for it!

I did have a case where ST failed to identify the split in the platen and thought I might mention it. Also, I tried to use the "manual" setting and apply to all pages, but I couldn't really tell the difference between manual and auto--in both cases, ST set the split line. I'd hoped to be able to set the split line myself manually in the first image, and then have ST use that same setting for all additional images (I'd selected to have it apply to all pages). I'm guessing that that must not be the function of manual. If not, I wonder, what is?

Thanks for a great program,
J

[daniel_reetz]

Hey jonEP, two thoughts... maybe this would go better in the Pre thread (easier for Tulon to find)... and I know Tulon really appreciates original sample pages that cause problems, including them would be a big help!

[spamsickle]

I think "manual" mode in "page split" only changes one page, and doesn't offer an "apply to" option that would change other pages. The "apply to" option just propagates the "hint" about how the page is split, not the actual coordinates of the split you specified.

In another post about a deskew problem, Tulon suggested manually changing one page, then editing the project file to apply that change to other pages. I've used that technique to apply a content selection to multiple pages, but I hadn't thought to use it for "split pages". I'd like to try it, because some of the scans I did before I knew about Scan Tailor (and which I'm going back and processing now) are getting more than 60% errors on the split page step, and those are usually the source of subsequent content selection errors as well. I'm worried that my "left" pages may not split the same way as my "right" pages, which would make editing the project file about as slow as going through and changing each page manually, but it may be that the two split lines are still in roughly the same place. If I do try it, I'll post results here.

And, if I ever get around to modifying Scan Tailor code itself, adding "apply to" for split pages will be right after adding "apply to" for content selection.

[Tulon]

Page splitting stage shouldn't make that many mistakes. Post a sample page if you want me to take a look.

[JonEP]

Thanks to you both, and continued thanks to Tulon

Wow, Spamsickle, if you ever got around to contributing a fixed content box to ST, and fixed split-pages, you would have a lot of grateful admirers!

Here's a sample tiff that didn't split (I've been having a lot of these--I imagine it's the reflection, although I'm not sure what I've done to my rig that it now creates a reflection that is more difficult for ST to interpret than previously.

N.B: I'd be happy for this to go in the Pre thread...

[univurshul]

x

[spamsickle]

A few of the books scanned with my first scanner cause the Page Split processing to make more mistakes than successes. It looks to me like reflections or misinterpreting book geometry accounts for most of these. Here's a screen shot of Scan Tailor's automatic page split results for a representative section:

PageSplit.jpg (73.01 KiB) Viewed 12174 times

For the page shown (and most right-hand pages), Scan Tailor appears to be identifying the reflection of the glossy book cover at the outer edge of the "target" page as its split line.

For the subsequent page (and most left-hand pages), it looks like the edge-on pages from the non-target side of the scanner are similarly confusing.

The originals can be found here.

I suspect I can crop out the parts of the image which are confusing Scan Tailor in a pre-processing step, but it may be possible to modify Page Split to handle such situations, if there is interest in doing so.

[Misty]

I'm curious. Does the page split algorithm have as much benefit to camera scans as for flatbed scans? My experience has typically been that pages manually set to have no splits process just as well as pages which were split, without incorrectly selecting anything outside the page for the content area. Since the page split algorithm does make mistakes sometimes, it seems to me that it can be more of a hindrance than a help for camera scanning.

[Tulon]

In both samples we have the same problem: they fail Scan Tailor's expectations. When you use "Page + offcut" mode, ST doesn't expect to see both vertical edges of the book. It expects to see the spine line and one or maybe even no vertical edges. You are giving it two. It's actually quite hard to tell a book's edge from its spine, so ST doesn't even try. In "Page + offcut" mode it does the following:
1. Find all more or less vertical lines.
2. Take the are between the leftmost and the rightmost of those. Cut off the rest.

My experience has typically been that pages manually set to have no splits process just as well as pages which were split, without incorrectly selecting anything outside the page for the content area.

By not cutting off the neighbouring page, you make the life of "Select Content" stage harder. It may or may not be able to cope with that.

[univurshul]

x