在使用 Tesseract 进行文本识别之前预处理图像

Question

I have a scanned page, that I'm trying to identify and parse the numbers from the image (line by line). In order to do that, I'm using Python Pytessarect, with the following code:

img = cv2.imread('image.jpg',0)
ret,thresh1 = cv2.threshold(img,110,255,cv2.THRESH_TOZERO)
scan_config = r'--oem 3 --psm 6'
extracted_text = pytesseract.image_to_string(thresh1, config=scan_config)

input image.jpg:

Unfortunately, the result is not satisfying because as you can see, the digits on the 4th column are partly erased (a human eye can identify the digits, but a threshold algorithm makes it even worse):

006442000180
006354924010
005900000461
062891556156
006*3*00000261
006900000261

Someone has an idea of how to pre-process the image so that the algorithm will be able to identify even the party erased digits? By the way, the 2nd argument of the threshold function is hardcoded (110) and it probably won't match every image out there, it depends on the photo's quality, is there a way to generate the value dynamically or to use alternative to the threshold approach (maybe using OpenCV filters)?

Answer 1

tesseract PzCox.png - --dpi 72 --psm 6

produce this with (English) model best :

006442000180
006354924010
005300000461
062891556156
006300000261
006300000261