표준화 된 이미지에서 테두리가있는 섹션을 확인하고 파이썬으로 자르기

Question

초보자를위한 사과 - 저는 진정한 파이썬 초보자입니다.

첨부 이미지는 UB-04 양식 (또는 CMS-1450)이며 청구 청구를 위해 기관 공급자가 사용하는 균일 한 청구서 양식입니다. Medicare and Medicaid (CMS)에 의해 The Centers에서 개발되었지만 모든 보험 회사에서 사용하는 표준 양식이되었습니다.

UB04 Example

나는 각 공급자들이 정부 나 민간 보험에 의한 지급을 받기 위해 준 치료에 관련하여 정보를 채웠다 PDF로이 양식을받을 수 있습니다. 필자는이 PDF에서 수동으로 정보를 추출했지만이 작업을보다 빠르게 진행할 수있는 프로세스를 만들었습니다. 그러나이 프로세스는 충분히 강력하지 않으며 가능하면 개선하기를 원합니다.

는 제가 임의로 편집 텍스트와이 양식의 샘플을 추가 한

#I need to clean up these imports -_- 
import pytesseract 
from PIL import Image as IMG 
import PIL 
import cv2 
from wand.image import Image 
import PythonMagick 
import pandas as pd 
import operator 
import functools 
import csv 
from openpyxl import Workbook, load_workbook 
import numpy as np 
import math 
from matplotlib import pyplot as plt 
import os 
import xlsxwriter 
from random import randint 
import glob 
import datetime 

# Choose PDF series to convert to images - parse the images page by page 
pdfclaimtoconvert = "UB04 PDF" 

with(Image(filename=pdfclaimtoconvert+".pdf",resolution=200)) as source: 
    images=source.sequence 
    pages=len(images) 
    for i in range(pages): 
     Image(images[i]).save(filename='Cropped Images/'+pdfclaimtoconvert+str(i)+'.tiff') 

# Loop through all pages 
for page in range(0,pages): 
    # Select page number to parse text from 
    pagenumber = str(page) 

    filetoworkon = 'Cropped Images/'+pdfclaimtoconvert+pagenumber+'.tiff' 

    # Read & Write image back as tiff 
    image = cv2.imread(filetoworkon) 
    cv2.imwrite("test.tiff",image) 

    # Convert Red Pixels to white (this helps eliminate noise for tesseract to work properly) 
    image_tiff = IMG.open('test.tiff') 
    # load the pixel info 
    width, height = image_tiff.size 
    for x in range(width): 
     for y in range(height): 
      r,g,b = image_tiff.getpixel((x,y)) 
      if r > 130 and b < 240: 
       image_tiff.putpixel((x, y), (255,255,255)) 
    image_tiff.save('test.tiff') 

이 이번 섹션은 지금 내 주요 세차게 내 던지다입니다 -. 이상적으로, 오히려 엄격하게 세그먼트를 지정하는 것보다, 내가 좋아하는 것 그 경계의 각 하나를 식별 할 수있는 프로세스를 시작합니다 (예를 들어 왼쪽 상단은 경계선 1, 그 옆에있는 경계선은 경계선 2 등입니다) -이 문서에는 수백 개가있을 것이라고 가정합니다.) 그런 다음 그 경계에서 정확히 구분할 스크립트를 실행하고 필요에 따라 정보를 구문 분석 할 수 있습니다.

#Read Image and Crop relevant sections 
    img = cv2.imread('test.tiff') 

    crop_img = img[20:170, 20:530] # Address 
    cv2.imwrite("Cropped Images/test1.tiff", crop_img) 

    crop_img = img[200:240, 20:530] #patient name 
    cv2.imwrite("Cropped Images/test2.tiff", crop_img) 

    crop_img = img[260:310, 20:205] #birthdate 
    cv2.imwrite("Cropped Images/test3.tiff", crop_img) 

    crop_img = img[400:570, 20:860] #Payer Address 
    cv2.imwrite("Cropped Images/test4.tiff", crop_img) 

    crop_img = img[600:1340, 20:120] #Treatment Codes 
    cv2.imwrite("Cropped Images/test5.tiff", crop_img) 

    crop_img = img[600:1340, 121:620] #Treatment Descriptions 
    cv2.imwrite("Cropped Images/test6.tiff", crop_img) 

    crop_img = img[600:1340, 620:910] #HCPCS 
    cv2.imwrite("Cropped Images/test7.tiff", crop_img) 

    crop_img = img[600:1340, 910:1059] #Service Dates 
    cv2.imwrite("Cropped Images/test8.tiff", crop_img) 

    crop_img = img[600:1340, 1059:1220] #Service Units 
    cv2.imwrite("Cropped Images/test9.tiff", crop_img) 

    crop_img = img[600:1340, 1214:1365] #Service Charges 
    cv2.imwrite("Cropped Images/test10.tiff", crop_img) 

    crop_img = img[600:1340, 1355:1420] #Service Charges decimals 
    cv2.imwrite("Cropped Images/test11.tiff", crop_img) 

    crop_img = img[1400:1510,20:480] #Payer Name 
    cv2.imwrite("Cropped Images/test12.tiff", crop_img) 

    crop_img = img[20:75,1070:1580] #Patient Control No 
    cv2.imwrite("Cropped Images/test13.tiff", crop_img) 

    crop_img = img[75:105,1070:1400] #Med Rec 
    cv2.imwrite("Cropped Images/test14.tiff", crop_img) 

    crop_img = img[130:175,1015:1220] #Fed Tax No 
    cv2.imwrite("Cropped Images/test15.tiff", crop_img) 

    crop_img = img[135:175,1220:1510] #Statement from and To 
    cv2.imwrite("Cropped Images/test16.tiff", crop_img) 

    crop_img = img[1340:1372,900:1050] #Creation Date 
    cv2.imwrite("Cropped Images/test17.tiff", crop_img) 

    crop_img = img[1662:1700,630:1160] #Document Control No 
    cv2.imwrite("Cropped Images/test18.tiff", crop_img) 

    crop_img = img[1340:1372,130:280] #Pages #1 
    cv2.imwrite("Cropped Images/test19.tiff", crop_img) 

    crop_img = img[1340:1372,280:500] #Pages #2 
    cv2.imwrite("Cropped Images/test20.tiff", crop_img) 

    # Use Tesseract to Read text for each of the crops - input in a list 
    # Separate numerical exrtactions from mixed or string so that we can force tesseract to recognize them as digits 
    TextExtract = {} 

    numericals = [1, 2, 3, 4, 5, 6, 7, 8, 12, 13, 14, 15, 16, 17, 18] 
    nonnumericals = [9, 10, 11, 19, 20] 

    for i in numericals: 
     img = IMG.open(
      'YOURPATH/Cropped Images/test' + str(i) + '.tiff') 
     img.load() 
     TextExtract["test{0}".format(i)] = pytesseract.image_to_string(img, config='-psm 6') 

    for i in nonnumericals: 
     img = IMG.open(
      'YOURPATH/Cropped Images/test' + str(i) + '.tiff') 
     img.load() 
     TextExtract["test{0}".format(i)] = pytesseract.image_to_string(img, 

                   config='-c tessedit_char_whitelist=-psm 6') 

    # Split each item in dictionary by page break 
    ParsedText = {} 

    for i in range(1, 21): 
     ParsedText["test{0}".format(i)] = TextExtract['test' + str(i)].split('\n') 

    # Delete empty list items 
    for x in range(1, 21): 
     ParsedText['test' + str(x)] = [i for i in ParsedText['test' + str(x)] if i != ''] 

    # Collapse lists into single values 
    CollapsedLists = {} 

    collapsablelist=(1,2,3,4,12,13,14,15,16,17,18,19,20) 

    for i in range(1,21): 
     if i in collapsablelist: 
      CollapsedLists["test{0}".format(i)] = ' '.join(ParsedText['test'+str(i)]) 
     else: 
      CollapsedLists["test{0}".format(i)] = ParsedText['test'+str(i)] 

    # Extraction list 
    extractionlist = (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20) 

    # Create empty Pandas Dataframe 
    extractionframe = pd.DataFrame(columns=['test1', 'test2', 'test3', 'test4', 'test5', 'test6', 'test7', 'test8' 
     , 'test9', 'test10', 'test11', 'test12', 'test13', 'test14', 'test15', 'test16' 
     , 'test17', 'test18', 'test19', 'test20' 
              ]) 

    # Populate dataframe 
    for x in extractionlist: 
     if isinstance(CollapsedLists['test' + str(x)], list) is False: 
      try: 
       extractionframe.loc[1, 'test' + str(x)] = CollapsedLists['test' + str(x)] 
      except: 
       pass 
     else: 
      for i in range(len(CollapsedLists['test' + str(x)])): 
       extractionframe.loc[i + 1, 'test' + str(x)] = CollapsedLists['test' + str(x)][i] 

    # Populate source 
    extractionframe.loc[1, 'source'] = pdfclaimtoconvert + '_page_' + pagenumber 

    # Fill NaN values forward 
    extractionframe = extractionframe.fillna(method='ffill') 

    #Convert to excel formula in order to preserve any leading zeroes 
    extractionframe = extractionframe.applymap(str) 
    extractionframe2 = '="'+extractionframe+'"' 

    # Export to excel 
    with open('Claim Data Extractions.csv', 'a') as f: 
     extractionframe2.to_csv(f, header=False, index=False) 

Answer 1

등고선 분석이 필요합니다. 대부분의 노력은 OpenCV에 캡슐화되어 있으며 여러분에게 영감을주는 훌륭한 예가 있습니다 here.