Search code examples
pythonopencvtensorflowtesseract

Extract number from image with 2D array using Python tesseract

During extract Integers from a image containing a 2D Matrix form Tesseract is unable to give the correct result and the result varies every time we execute the code can anybody please give some idea what is missing from the code below 

img = cv2.imread(img_path)
rows = img.shape[0]
cols = img.shape[1]

#print rows , cols

img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

# Apply dilation and erosion to remove some noise
#kernel = np.ones((5,5), np.uint64)
#img = cv2.dilate(img, kernel, iterations=1)
#img = cv2.erode(img, kernel, iterations=1)

# Write image after removed noise

cv2.imwrite(src_path + "removed_noise1.png", img)

#  Apply threshold to get image with only black and white
#img = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY,225,95)
img = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C , cv2.THRESH_BINARY ,251,95)
#print cv2.getGaussianKernel(ksize=221,sigma=41)

# Write the image after apply opencv to do some ...
cv2.imwrite(src_path + "thres1.png", img)

# Recognize text with tesseract for python  
result = pytesseract.image_to_string(Image.open(src_path + "thres1.png"))

Input

enter image description here

Threshold value : adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C , cv2.THRESH_BINARY ,251,95

Output of threshold :
enter image description here

Output is in the form of 1 5 5 7 5 7 3 8 6 4 9 0 2 4 8 6 1 3 0 2 3 9 0 8 9 can be either in the row major form and column major form doesn't matter but we do need the given output saved into a variable

Solution

  • Try changing the threshold value from 251, 95 to something like 251, 40. 

    img = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C , cv2.THRESH_BINARY ,251,40)

    It seems you already have perfect image which doesn't require any changes to extract string from image. Pytesseract's image_to_string doesn't work on my system so I used someone's made word by word OCR. This is definitely not the best solution in the world by hey if it works, it works. I have attached few files (see google drive links below)

    Steps:

    1. Run slice_img.py : this will break your image into 5x5 pieces.
    2. Run TrainAndTest.py : this will extract numbers from each slice of image we have created and save it to data.txt file.

    Please note:

    1. Training_chars.png, classifications.txt and flattened_images.txt are files necessary for OCR to work.
    2. You must have numpy, PIL, OpenCV libraries installed for the above script to work
    3. Please keep all files provided in same folder and use image name as test.png

    TrainAndTest.py - https://drive.google.com/file/d/0B05aeuFExe2Aa3p3SWszN2xqU2c/view?usp=sharing
    slice_image.py - https://drive.google.com/file/d/0B05aeuFExe2AN0t3UUlGZ3VjcW8/view?usp=sharing
    training_chars.png - https://drive.google.com/file/d/0B05aeuFExe2ANjJNbzV5VTJyRTA/view?usp=sharing
    classifications.txt - https://drive.google.com/file/d/0B05aeuFExe2AZU91bUpOblB3d2c/view?usp=sharing
    flattened_images.txt - https://drive.google.com/file/d/0B05aeuFExe2AeXVnbXVXVTZ2RTQ/view?usp=sharing