Extract Hand written text on second page of multiple PDF files using python

Question

I am beginner in python and not sure how to go about achieving this task. I have several scanned PDF files in a folder. The hand written text that I am interested in extracting is in the second page of each of those PDF files. I am trying to follow the instructions given in this blog: https://blog.hexstream.com/process-handwritten-text-using-python-and-cloud-vision

But I am totally lost how to lay it all out in code where it should iterate through all the pdf files in the folder, extract the hand written text on the second page and store that in a dataframe where one column will be the PDF file name and the second colum will have the hand written text. Finally to output the datafame to an excel or csv file. Any help would be much appreciated.

Answer 1

You can use pypdf2 and ocropus library. It is complicated to explain in one place or to write a function but refer to the documentation of these libraries and you will get to know what to do and how to implement. Below are some general steps.

1. extract writing from image using ocropus library
2. extract handwritten text
3. finally get read all files from dir

Here is a short implementation.

def extract_handwriting_from_pdf(pdf_path):

with open(pdf_path, 'rb') as f:

pdf_reader = PdfFileReader(f)

if pdf_reader.numPages < 2:

return ''

page = pdf_reader.getPage(1)

xObject = page['/Resources']['/XObject'].getObject()

for obj in xObject:

if xObject[obj]['/Subtype'] == '/Image':

size = (xObject[obj]['/Width'], xObject[obj]['/Height'])

data = xObject[obj].getData()

image = Image.frombytes('L', size, data)

text = extract_handwriting(image)

if handwriting_pattern.match(text):

return text

return ''

This is just an example please adjust to your use case accordingly. Great way is to read documentation of the above libraries and you will get to know as to how to do this. Good luck.

Answer 2

The second step of the tutorial is where it takes one pdf file, and splits each page into an image. The third step is where the code iterates through all the pages of the pdf. Since you do not want all the pages, only the second, you do not need to iterate. Looking at the documentation, the convert_from_path function creates a list of images. You only want the second image, so images[1] should get you the second page. (images is the variable assigned to output of the convert_from path call).

I believe the tutorial goes over how to store it in a dataframe. If you want to save it as an excel or csv, pandas (I am assuming you are working with pandas dataframes) has two useful functions for you.

to_csv and to_excel

something like dataframe_name.to_csv("output.csv") should be what you are looking for. I just saw that the tutorial goes over this step as the last line of code in step 14,

The last thing would be doing this for multiple pdf files. If the filenames of the pdf files are predictable, you should use python to create those file names, use a for loop to iterate over them, and manually check and create the stopping condition, ie if its 1.pdf, 2.pdf etc, create a counter, convert it to string, then add ".pdf" to the end, manually check how many files to know how many times to do this. Another approach could be to use os.listdir to first get all the file names then create a for loop to iterate over all the file names. This goes before the rest of your code. The reason I explained it last, is because you should make sure your code works for one specific pdf or if you want even before that one specific image you manually converted, before putting it into this loop. The rest of your code goes inside this for loop.

The documentation for most of these functions is pretty good, and python is pretty good at telling you which line is messing up, so referencing that is always useful.

Also I noticed the tutorial is asking you to use credentials and access an online resource Google Cloud Visions. You are batch processing images, as in doing this multiple times. You computer and Googles computers might be faster than the internet, and this sometimes causes problems. Some on-line resources don't like you accessing them so fast as well, while the functions you are using most likely have some way to handle that, beware your code might be correct, but through up errors because Google is being finicky.

Best of luck to you.