Python Tesseract 11 segmentation defect

Question

I use tesseract and opencv to read images as part of another program. I wrote this play to try to work out the errors that I have with the main program. The test () function will be largely copied and pasted into the final program. The problem I'm running into is that tesseract seems to be shutting down with python using segmentation error 11. This happens periodically. Sometimes this sample code works completely, sometimes it fails after 56 iterations (each time). Everything that I installed is in homegrown (I followed the following instructions: https://code.google.com/p/python-tesseract/wiki/HowToCompilePythonTesseractForHomebrewMacMountainLion ) After finding solutions to my problem, I tried the instructions in this: http: // www.janeriksolem.net/2011/12/installing-opencv-python-interface-on.html and I'm sure I did everything right, but I still get separate crashes.

import time import tesseract import cv2 import cv2.cv as cv import Image def test(): image0=cv2.imread("test.jpg") offset=20 height,width,channel = image0.shape image1=cv2.copyMakeBorder(image0,offset,offset,offset,offset,cv2.BORDER_CONSTANT,value=(255,255,255)) api = tesseract.TessBaseAPI() api.Init(".","eng",tesseract.OEM_DEFAULT) api.SetPageSegMode(tesseract.PSM_AUTO) height1,width1,channel1=image1.shape print image1.shape print image1.dtype.itemsize width_step = width*image1.dtype.itemsize print width_step #method 1 iplimage = cv.CreateImageHeader((width1,height1), cv.IPL_DEPTH_8U, channel1) cv.SetData(iplimage, image1.tostring(),image1.dtype.itemsize * channel1 * (width1)) tesseract.SetCvImage(iplimage,api) text=api.GetUTF8Text() conf=api.MeanTextConf() image=None print "..............." return (text, conf) for x in xrange(200): print "x: %d" %x test() print time.sleep(1) print "Done" 

Answer 1

I know this is a very late answer, but I had the same problem when working with OS X Lion, python2.7, OpenCV 2.4 and Tesseract 3.0

I found three problems when debugging using eclipse + pydev ...

First, the image given by tesseract should always be a binary image.

The second problem: cv2.imread() does not work sometimes, and it just returns an empty ndarray. I really don’t know why this is happening, and if it is related to running python in eclipse + pydev.

So when I tried to use empty ndarray shape and dtype , I had null characters, and when using cv.SetData() this created an empty iplimage. And when api.GetUTF8Text() tried to work on this empty iplimage, everything broke in a weird way, and I got "Segmentation Error 11"

Thirdly, it turns out that cv2 and cv have a completely different way of handling arrays, and they reset their axes. So, if you did something like ...

 image = cv2.imread('something.jpg',0) # flag = 0 is for converting to grayscale image = cv2.threshold(image,128,255,cv2.THRESH_BINARY) height,width,channel = image.shape 

then you need to do ...

 iplimage = cv.CreateImageHeader((width,height), cv.IPL_DEPTH_8U, 1) cv.SetData(iplimage, image.tostring(),image.dtype.itemsize * (width)) #the number of channels is 1 

and I see that you have solved the third problem.

It's strange if I have a code like img0 = cv.fromarray(scr) right after the line scr = cv2.imread('textSample.jpg',0) , then everything works fine.

Go through the following code with the image "textSample.jpg" (at the bottom of this post) and uncomment the lines (except the first line, which is not a comment), so you can see which image your codes are working on. scr , img0 and img1 should ultimately be the same:

 #!/usr/bin/env python import cv2 import cv2.cv as cv import tesseract scr = cv2.imread('textSample.jpg',0) #img0 = cv.fromarray(scr) #cv.SaveImage('img0.jpg',img0) api = tesseract.TessBaseAPI() api.Init(".","eng",tesseract.OEM_DEFAULT) api.SetPageSegMode(tesseract.PSM_AUTO) image = cv.CreateImageHeader((scr.shape[1],scr.shape[0]), cv.IPL_DEPTH_8U, 1) cv.SetData(image, scr.tostring(), scr.dtype.itemsize*scr.shape[1]) #cv.SaveImage('img1.jpg',image) tesseract.SetCvImage(image,api) text=api.GetUTF8Text() conf=api.MeanTextConf() print text print conf 

you should get something like ...

 OE3456789 !"#$%&'()* .-./ 76