Getting the bounding box of the recognized words using python-tesseract

Question

I am using python-tesseract to extract words from an image. This is a python wrapper for tesseract which is an OCR code.

I am using the following code for getting the words:

import tesseract

api = tesseract.TessBaseAPI()
api.Init(".","eng",tesseract.OEM_DEFAULT)
api.SetVariable("tessedit_char_whitelist", "0123456789abcdefghijklmnopqrstuvwxyz")
api.SetPageSegMode(tesseract.PSM_AUTO)

mImgFile = "test.jpg"
mBuffer=open(mImgFile,"rb").read()
result = tesseract.ProcessPagesBuffer(mBuffer,len(mBuffer),api)
print "result(ProcessPagesBuffer)=",result

This returns only the words and not their location/size/orientation (or in other words a bounding box containing them) in the image. I was wondering if there is any way to get that as well

may help https://github.com/tesseract-ocr/tesseract/wiki/APIExample#result-iterator-example — iMath, Dec 24 '18 at 07:53

stwykd · Accepted Answer · 2019-01-06T14:50:18.020

125

Use pytesseract.image_to_data()

import pytesseract
from pytesseract import Output
import cv2
img = cv2.imread('image.jpg')

d = pytesseract.image_to_data(img, output_type=Output.DICT)
n_boxes = len(d['level'])
for i in range(n_boxes):
    (x, y, w, h) = (d['left'][i], d['top'][i], d['width'][i], d['height'][i])
    cv2.rectangle(img, (x, y), (x + w, y + h), (0, 255, 0), 2)

cv2.imshow('img', img)
cv2.waitKey(0)

Among the data returned by pytesseract.image_to_data():

left is the distance from the upper-left corner of the bounding box, to the left border of the image.
top is the distance from the upper-left corner of the bounding box, to the top border of the image.
width and height are the width and height of the bounding box.
conf is the model's confidence for the prediction for the word within that bounding box. If conf is -1, that means that the corresponding bounding box contains a block of text, rather than just a single word.

The bounding boxes returned by pytesseract.image_to_boxes() enclose letters so I believe pytesseract.image_to_data() is what you're looking for.

edited Jan 06 '19 at 14:50

answered Jan 06 '19 at 06:23

stwykd

2,684
3
19
20

2

This is actually the correct answer for this question. But might be ignored by people due to complexity of this method – Parikshit Chalke Jan 25 '19 at 06:59
1

Do you know the meaning of other columns(level, page_num, block_num, par_num, line_num, word_num) in the output generated by image_to_data? – Eswar RDS Apr 27 '20 at 12:58
1

This work only for tesseract >= 3.05. I need a solution for lower version. – Bùi Nhật Duy Jun 16 '20 at 16:06
When I ran this, the output from `pytesseract` showed that basically the whole image was the bounding box: https://i.imgur.com/E9dKc3L.png – Raleigh L. Nov 07 '22 at 09:00

score 20 · Answer 2 · answered Dec 30 '13 at 02:18

20

tesseract.GetBoxText() method returns the exact position of each character in an array.

Besides, there is a command line option tesseract test.jpg result hocr that will generate a result.html file with each recognized word's coordinates in it. But I'm not sure whether it can be called through python script.

answered Dec 30 '13 at 02:18

lennon310

12,503
11
43
61

5

I get result.hocr file with the command, though the file is HTML format. – Henry Jan 26 '17 at 00:59

score 16 · Answer 3 · answered Apr 20 '18 at 12:16

16

Python tesseract can do this without writing to file, using the image_to_boxes function:

import cv2
import pytesseract

filename = 'image.png'

# read the image and get the dimensions
img = cv2.imread(filename)
h, w, _ = img.shape # assumes color image

# run tesseract, returning the bounding boxes
boxes = pytesseract.image_to_boxes(img) # also include any config options you use

# draw the bounding boxes on the image
for b in boxes.splitlines():
    b = b.split(' ')
    img = cv2.rectangle(img, (int(b[1]), h - int(b[2])), (int(b[3]), h - int(b[4])), (0, 255, 0), 2)

# show annotated image and wait for keypress
cv2.imshow(filename, img)
cv2.waitKey(0)

answered Apr 20 '18 at 12:16

jtbr

1,149
13
13

Why `y-coordinate` is subtracted from the height of the image – Atinesh Feb 19 '19 at 09:47
2

I believe the pytesseract and opencv have different notions of the origin of the image (top left or bottom left), or at least that's what I I seemed to experience when I wrote the answer. If it works without the h there, great. – jtbr Apr 12 '19 at 17:08

score 7 · Answer 4 · answered Jul 13 '17 at 11:45

Using the below code you can get the bounding box corresponding to each character.

import csv
import cv2
from pytesseract import pytesseract as pt

pt.run_tesseract('bw.png', 'output', lang=None, boxes=True, config="hocr")

# To read the coordinates
boxes = []
with open('output.box', 'rb') as f:
    reader = csv.reader(f, delimiter = ' ')
    for row in reader:
        if(len(row)==6):
            boxes.append(row)

# Draw the bounding box
img = cv2.imread('bw.png')
h, w, _ = img.shape
for b in boxes:
    img = cv2.rectangle(img,(int(b[1]),h-int(b[2])),(int(b[3]),h-int(b[4])),(255,0,0),2)

cv2.imshow('output',img)

doesn't work, boxes is unknown parameter in lastest pytesseract — Stepan Yakovenko, Dec 13 '18 at 08:44

score 4 · Answer 5 · answered Oct 18 '18 at 18:21

Would comment under lennon310 but don't have enough reputation to comment...

To run his command line command tesseract test.jpg result hocr in a python script:

from subprocess import check_call

tesseractParams = ['tesseract', 'test.jpg', 'result', 'hocr']
check_call(tesseractParams)

score 4 · Answer 6 · edited Jul 19 '23 at 02:29

To get bounding boxes over words:

import cv2
import pytesseract
from pytesseract import Output

img = cv2.imread('test-01.jpg')
d = pytesseract.image_to_data(img, output_type=Output.DICT)
n_boxes = len(d['level'])
for i in range(n_boxes):
    if (d['text'][i] != ""):
        (x, y, w, h) = (d['left'][i], d['top'][i], d['width'][i], d['height'][i])
        cv2.rectangle(img, (x, y), (x + w, y + h), (0, 255, 0), 2) # in-place operation

cv2.imwrite('result.png', img)

score 2 · Answer 7 · edited Sep 20 '22 at 12:23

Some examples are answered aove which can be used with pytesseract, however to use tesserocr python library you can use code given below to find individual word and their bounding boxes:-

    with PyTessBaseAPI(psm=6, oem=1) as api:
            level = RIL.WORD
            api.SetImageFile(imagePath)
            api.Recognize()
            ri = api.GetIterator()
            while True::
                word = ri.GetUTF8Text(level)
                boxes = ri.BoundingBox(level)
                print(word,"word")
                print(boxes,"coords")
                if not ri.Next(level):
                     break

score 0 · Answer 8 · answered Jan 16 '22 at 20:51

0

As already mentioned, you can use pytesseract's image_to_boxes. You can check my Docker Hub repo https://hub.docker.com/r/milanhlinak/tesseract-image-to-boxes - a simple Flask application with Tesseract 5.0.0.

answered Jan 16 '22 at 20:51

Milan Hlinák

4,260
1
30
41

Getting the bounding box of the recognized words using python-tesseract

8 Answers8

Linked