Segfault in pixa_to_list

[oskaritimperi]

Hi, I've encountered a segfault in pixa_to_list and I can reproduce it consistently. I don't have any idea how to fix this though.

This image here will always make tesserocr segfault:

This image on the other hand works fine:

The code I'm using for testing is simple:

import tesserocr
from PIL import Image
import sys

print(tesserocr.tesseract_version())
print(tesserocr.get_languages())

png = Image.open(sys.argv[1]).convert('L')

# print(tesserocr.image_to_text(png))

with tesserocr.PyTessBaseAPI() as api:
    api.SetImage(png)
    boxes = api.GetComponentImages(tesserocr.RIL.WORD, True)
    for _, box, _, _ in boxes:
        pad = box['h'] * 0.2
        api.SetRectangle(box['x']-pad, box['y']-pad, box['w']+pad, box['h']+pad)
        text = api.GetUTF8Text().strip()
        confidence=api.MeanTextConf()
        print(text, confidence)

Here is a crash report from OS X: crashreport.txt

Here is the output from a succesful run (including version numbers and so on): success.txt

I'm using tesseract 3.05.00 which I compiled myself as I had this problem with the 3.04 also and I thought maybe the new version would fix the issue.

Here are the relevant environment variables I used when I executed python setup.py install for tesserocr:

declare -x CFLAGS="-g -fno-omit-frame-pointer  -UNDEBUG -O0"
declare -x CPPFLAGS="-I/Users/otimpe/dev/tesseract-3.05.00/dist/include"
declare -x DYLD_LIBRARY_PATH="/Users/otimpe/dev/tesseract-3.05.00/dist/lib"
declare -x LDFLAGS="-L/Users/otimpe/dev/tesseract-3.05.00/dist/lib -g"
declare -x TESSDATA_PREFIX="/usr/local/share"

[sirfz]

Hi, this happened because GetComponentImages is returning a NULL boxa for this image. I wasn't validating this which caused pixa_to_list to run against an invalid pointer. I've added a check for this in all instances where pixa_to_list is used to ensure an empty list is returned instead of segfaulting. Thanks for the catch :)

You can test this using the tesseract4 branch.

[oskaritimperi]

Hi,

the fixes you made seem to work nicely with tesseract 3.04. :-) Thank you!