OCR application for invoices, built in Python
Budget: $30 – $250 USD
We want to develop a web application that uses free Python modules that can benefit of the GPU processing power to extract data from invoices (preferably Tesseract OCR, OpenCV and Python) and a web interface for it, with user and password.
Users should be able to upload their invoice one by one or more at a time in pdf/png/jpg format, even multipage, review the extracted data and then choose to export it in xml/csv/json format.
Invoices can contain one or more products (information lines) and for each line the application must contain a function of associating the accounting accounts based on the content / name of the item in the invoice (for example for apples we have account `1` and for cucumbers account `2`).
The records will be stored in a MySQL database, with filtering for each user and the invoice separately.
For newly document templates there must be the possibility of machine learning combined with the user ability to train it (for example to set the area for the invoice number and so on).
After processing, the application must highlight the bills to which the total amount calculated by adding the positions differs from the total inscribed in the invoice.
If there are invoices with erroneous values, the user must be able to manually correct those values, having in one part of the screen the records from the database and in another part the scanned invoice (maybe with a zoom function in the pointer area).
We will provide the database structure (SQL), links to other similar applications and as many and as many graphical representations as necessary.
Users should be able to upload their invoice one by one or more at a time in pdf/png/jpg format, even multipage, review the extracted data and then choose to export it in xml/csv/json format.
Invoices can contain one or more products (information lines) and for each line the application must contain a function of associating the accounting accounts based on the content / name of the item in the invoice (for example for apples we have account `1` and for cucumbers account `2`).
The records will be stored in a MySQL database, with filtering for each user and the invoice separately.
For newly document templates there must be the possibility of machine learning combined with the user ability to train it (for example to set the area for the invoice number and so on).
After processing, the application must highlight the bills to which the total amount calculated by adding the positions differs from the total inscribed in the invoice.
If there are invoices with erroneous values, the user must be able to manually correct those values, having in one part of the screen the records from the database and in another part the scanned invoice (maybe with a zoom function in the pointer area).
We will provide the database structure (SQL), links to other similar applications and as many and as many graphical representations as necessary.