ExtractWithTesseract.py

# -*- coding: utf-8 -*-
"""
Created on Thu Apr  5 17:23:53 2018
@author: ADubey4
"""

import cv2
from pytesseract import pytesseract as pt
import pandas as pd
from PIL import Image
import sys
import pdf2image
import numpy as np
from signatureExtractor import getSignatureFromPage, getSignature
#from matplotlib import pyplot as plt

if sys.version_info[0] < 3: 
    from StringIO import StringIO
else:
    from io import StringIO



pt.tesseract_cmd = r'C:\Program Files (x86)\Tesseract-OCR\tesseract'
file = r"C:\Users\file\path"


#search_word = "sincerely"
search_word = "signature"

signature_margin = (400, 400, 400, 400) # (left, top, right, bottom)
is_signature_margin_percentage = True # if true its percentage of the box dimensions, else the exact pixels


if(is_signature_margin_percentage):
    signature_margin = [x/100.0 for x in signature_margin ]

if file.endswith(".pdf"):
    doc_page_images = pdf2image.convert_from_path(file, dpi = 300)
else:
    img = cv2.imread(file,cv2.COLOR_RGB2BGR)
    doc_page_images=[img]
#    img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

def get_ocr_df(doc):
    ret = pt.image_to_data(doc).lower()
    TESTDATA = StringIO(ret)
    return pd.read_csv(TESTDATA, sep=r"\t", engine='python').dropna(axis=0, how='any')

def clean_printed(im):
    df = get_ocr_df(im)
    print(df)
    for i, row in df.iterrows():
        im[row.top:row.top + row.height+1, row.left:row.left + row.width+1] = 255
    return im

for page_no, doc in enumerate(doc_page_images):
    df = get_ocr_df(doc)
    signature_df = df[df['text'].str.contains(search_word)]

    if type(doc) is np.ndarray:
        x2_max = doc.shape[0]
        y2_max = doc.shape[1]
        doc_type = "np"
    else:
        x2_max = doc.size[0]
        y2_max = doc.size[1]
        doc_type = "pil"

    for i, row in signature_df.iterrows():
        print(page_no, i)
        if is_signature_margin_percentage :
            x1 = row.left - int(signature_margin[0]*row.width)
            y1 = row.top - int(signature_margin[1]*row.height)
            x2 = row.left + row.width + int(signature_margin[2]*row.width)
            y2 = row.top + row.height + int(signature_margin[3]*row.height)
        else:
            x1 = row.left - int(signature_margin[0])
            y1 = row.top - int(signature_margin[1])
            x2 = row.left + row.width + int(signature_margin[2])
            y2 = row.top + row.height + int(signature_margin[3])
        
        if x1 < 0: x1=0
        if y1 < 0: y1=0
        if x2 > x2_max: x2 = x2_max
        if y2 > y2_max: x2 = y2_max
        
        print(x1,y1, x2,y2)
        if type(doc) is np.ndarray:
            im = doc[y1:y2+1, x1:x2+1]
            Image.fromarray(im).show()
        else:
            im = doc.crop((x1, y1, x2, y2))
            im.show()
        
        im = clean_printed(np.array(im))
        Image.fromarray(im).show()
#        signature = getSignatureFromPage(img = np.array(im))
#        signature = getSignature(img = signature)
#        cv2.imshow('Signature'+str(i), signature)