Say there is a set of scanned documents one wants to classify. One can assume there is a set of manually selected examples for each class.
The classes would contain, for example: receipts, medical reports and invoices.
My go-to algorithm would be K-NN, but the question would be: which features to use?
Thiqs isq a good question.
Here are features to consider:
skew:
https://www.researchgate.net/publication/289093892_Skew_Detection_of_Scanned_Document_Images
https://www.researchgate.net/publication/307516380_Signature_line_detection_in_scanned_documents
Article Skew Detection of Scanned Document Images
Conference Paper Signature line detection in scanned documents
Thank you James for your answer. I reckon that skew and signature are good features if you want to identify documents from the same persons or scanned using the same procedure. However, I need to cluster together documents of the same type edited by different persons. For example, formA edited by personX together with formA edited by person.
To begin with, I thought about using the following:
- number of white rows
- number of white lines
What do you think?
Classification can be done by specifying pre-defined set of keywords and creating its databases.
then with the help of mapping technique you can map extracted words of the documents and then classify it.
- Badges
- Science topic