This project focuses on the extraction of Names, PAN ID and Aadhaar ID from a scanned document like pdf, images, etc using OCR (Optical Character Recognition) for extraction of text from images, pre-trained Hugging Face NER pipeline model for name extraction and regex module for pattern matching of the PAN cards and Aadhaar Card. The output is a dictionary containing Names and ID's associated with the given name.
code_files
| names_ner.py
| ocr.py
| output_phrasing.py
| re_pan_aadhaar_match.py
| result.py
Dataset
| (Images for testing)
README.md
It contains two function:
To read and extract the text from images using the Pytesseract OCR and Cv2.
def text_image(path:str)->str:
# Reading images using cv2 module and converting the
# extracted image to grayscale for uniformity for
# improved text extraction
image = cv2.imread(path)
image = cv2.cvtColor(image,cv2.COLOR_RGB2GRAY)
# Using Pytesseract for converting image to string
ocr_text = pytesseract.image_to_string(image,output_type=Output.STRING)
# Text processing
result = text_process(ocr_text)
return result
To process the extracted text to remove the extra spaces and indentations.
def text_process(text:str)->str:
# Removing the extra spaces and new line from the
# extracted text and converting into upper case
ocr_text = text.lstrip().rstrip().upper()
result = re.sub("\n+"," ",ocr_text)
result = re.sub("\s+"," ",result)
return result
It contains two function:
To extract the matching pattern of the PAN card from image text extracted using text_image function using the regex module.
def extract_PAN(text:str)->list:
# Defining pattern for PAN Card
pattern = "[A-Z]{3}[ABCFGHLJPTF][A-Z][0-9]{4}[A-Z]"
# Converting the text to upper case
text = text.upper()
# Finding the PAN Card ID using above pattern
result = re.findall(pattern,text)
# If the pattern is not found an empty list is returned
if len(result) == 0:
return []
return result
To extract the matching pattern of the extract_aadhaar card from image text extracted using text_image function using the regex module.
def extract_Aadhaar(text:str)->list:
# Defining pattern for Aadhaar Card
pattern = "\d{4}-\d{4}-\d{4}|\d{4} \d{4} \d{4}"
# Finding the Aadhaar Card numbers using above pattern
result = re.findall(pattern,text)
if len(result) == 0:
return []
return result
It contains a single function extract_names that takes string as input and returns the list of the names extracted using a pre-trained Hugging Face NER model. The model used is "dslim/bert-base-NER-uncased" which is trained on a rich dataset.
Only the names of the person represented as 'NER' is returned as per the requirement.
def extract_names(text:str)->list:
# Using a pre-trained Hugging Face model for extracting the names
# by using the NER pipeline
# Loading the tokenizer
tokenizer = AutoTokenizer.from_pretrained("dslim/bert-base-NER-uncased")
# Loading the model
model = AutoModelForTokenClassification.from_pretrained("dslim/bert-base-NER-uncased")
# Defining the NER pipeline with entities as grouped
nlp = pipeline("ner", model=model, tokenizer=tokenizer,grouped_entities=True)
# Extracting the names
ner_results = nlp(text)
names_list = []
# Extracting only the Person name defined as 'PER' by the model
for obs in ner_results:
if obs['entity_group'] == 'PER' and len(obs['word']) > 5:
names_list.append(obs['word'])
return names_list
This file contains function final_output. This function converts the output into the dictionary for the output to the result.py file.
def final_output(names:list,pan_id:list,aadhar_id:list)->list:
sol = []
# If the Aadhaar ID and PAN ID are empty then returning the list
# with NO ID's
if len(pan_id) == 0 and len(aadhar_id) == 0:
sol.append("No ID found")
# If the Aadhaar ID is empty then returning the list
# with Names and PAN ID
elif len(aadhar_id) == 0:
for name,pan in zip(names,pan_id):
zipped = {}
zipped["Name"] = name
zipped["PAN"] = pan
zipped["Aadhar"] = ""
sol.append(zipped)
# If the PAN ID is empty then returning the list
# with Names and Aadhaar ID
elif len(pan_id) == 0:
for name,adh in zip(names,aadhar_id):
zipped = {}
zipped["Name"] = name
zipped["PAN"] = ""
zipped["Aadhar"] = adh
sol.append(zipped)
# If the Aadhaar ID and PAN ID are not empty then returning the list
# with Names, PAN and Aadhaar ID's
else:
for name,pan in zip(names,pan_id):
zipped = {}
zipped["Name"] = name
zipped["PAN"] = pan
zipped["Aadhar"] = ""
sol.append(zipped)
return sol
This file compiles functions of all the above files and prints the output in the dictionary format.
def extract(path:str):
# Calling the text_image function in ocr file for text extraction
text = text_image(path)
# Calling the extract_names, extract_PAN and extract_Aadhaar functions
# in names_ner and re_pan_aadhar_match files respectively for
# data extraction
names = extract_names(text)
pan_id = extract_PAN(text)
aadhaar_id = extract_Aadhaar(text)
# Calling the final_output function in output_phrasing file for
# phrasing output text extracted above
sol = final_output(names,pan_id,aadhaar_id)
# Printing the the output
print(sol)
if __name__ == '__main__':
path = input("Enter the path of the image date to be extracted: ")
extract(path)
Server: Python, Visual Studio Code, Pytesseract, Hugging Face Transformers, Regex
I'm an ambitious and hardworking pre-final year student, with skills in Machine Learning, NLP, Data Science, Analytics, and Development.
Python, C++, Deep Learning, Image Processing, Flask, Heroku, NLP (Natural Language Processing), Machine Learning, Hugging Face, Image Processing, OCR(Optical Character Recognition), Regex
If you have any feedback, please reach out at rachitjindal56@gmail.com