pixGenHalftoneMask error

[scottb89]

(I'm using head: 3.05.00dev)

When ocr'ing a small one or two line image (with the default settings), I'm getting the following error:
Error in pixGenHalftoneMask: pix too small: w = 610, h = 71

But the ocr is successful.

In textord/imagefind.cpp, FindImages is calling pixGenHalftoneMask (a leptonica function) which is throwing the error and returning null.

FindImages is then executing it's backup plan, which works fine (such a small image doesn't need multiple images regions). It's backup plan is to call pixCreate and then return the result.

The minimum size for pixGenHalftoneMask to work is 100x100 pixels. Perhaps FindImages should
just go with plan 'b' if the size of the image is too small. This would still work fine and avoid throwing the error. I don't know if there is a way for teseract to know what the minimum size is (in case it changes).

[scottb89]

The min size is defined like this at the top of src/pageseg.c in leptonica:

/* These functions are not intended to work on very low-res images */

static const l_int32 MinWidth = 100;
static const l_int32 MinHeight = 100;

It doesn't seem that they are likely to change.

Here's my patch:

index 05047ca..0084ac8 100644
--- a/textord/imagefind.cpp
+++ b/textord/imagefind.cpp
@@ -72,6 +72,12 @@ Pix* ImageFind::FindImages(Pix* pix) {
pixDisplayWrite(pixr, textord_tabfind_show_images);

// Get the halftone mask directly from Leptonica.

• if (pixGetWidth(pixr) < 100 || pixGetHeight(pixr) < 100)
• {
• // leptonica will throw and error and return null
• pixDestroy(&pixr);
• return pixCreate(pixGetWidth(pix), pixGetHeight(pix), 1);
• }
  l_int32 ht_found = 0;
  Pix *pixht2 = pixGenHalftoneMask(pixr, NULL, &ht_found,
  textord_tabfind_show_images);

[jbreiden]

An alternative is for Leptonica to not spam stderr. I'm honestly not sure which approach is better.

[scottb89]

Good point. It's returning null, and the null is being handled properly. Automatically emitting the reason for the error should be directable/configurable.

I just dug into it and found NO_CONSOLE_IO referenced from environ.h.
Maybe I needed to turn that on when I built Leptonica. Still a questionable approach.

[amitdo]

@DanBloomberg, any comment?

[DanBloomberg]

Scott's patch seems reasonable to me.

Testing the size in advance prevents leptonica from returning null,
and from saying that an error occurred.

Slightly simpler just to
return pixCopy(pixr, pix);
which works whether or not pixr exists.

re: musings about whether or not leptonica should "spam stderr":

Just a reminder that leptonica provides a significant amount of control over messages to stderr, both at compile time and at run time, and broken down to the granularity of errors, warnings and information. The question for the leptonica client (tesseract, here) is what level of messages they want to receive on stderr, and in this case it's up to the maintainers. If the error messages (at least) are not on, a user will have no idea why something failed if it was due to bad input to a leptonica function, for example. On the other hand, when failures occur but are caught and handled by the client, the user may be unnecessarily concerned, when things are actually going ok.

FWIW, when I am programming and testing, I always use the default MINIMUM_SEVERITY of L_SEVERITY_INFO, which turns all messages on.

[amitdo]

OK. Thanks for your reply, Dan.

Just a quote:

To spam, or not to spam, that is the question.

;-)