-
-
input-invoice
-
output-1
-
output-2
Inspiration
We were really interested in learning Natural Language Processing and to mine interesting data on invoices. This can be used in real-life recommendation systems which might result in decreasing attrition rates and increasing sales.
What it does
The app takes in an invoice image and by a graph based OCR approach mines and extracts the data into an output csv file.
How we built it
The solution app is based on python. PyTesseract is used to OCR over invoice image files and levenshtein distance is used to get a sense on string sequences. The extracted data is treated as a graph and neighbor blocks are considered as potential key value pairs. The output is stored in csv format as the result.
Challenges we ran into
- As invoices do not follow a pattern, searching for a method to extract key value pairs is very difficult.
- Lack of time to explore the graph neural network modelling, which could have helped in getting a better distance approximation.
- Too sharp or blurred images resulted in worse output which was unexpected.
Accomplishments that we're proud of
- Coming up with the graph based neighbor approach to bring about correlation between two or more identified blocks.
What we learned
How to implement NLP and utilize graph based analysis over OCR data on images.
Log in or sign up for Devpost to join the conversation.