Simple-receipt-json-parser

Description

This is a pretty basic skeleton for a receipt parser. The structure allows you to add your custom rule-based or perhaps neural network-based parsers as well as implement a custom pre-processing pipeline for your images.

The receipt_parser.py script compiles the basic usage of the classes into a script which is designed to work with rather higher quality scans with 300 DPI ideally. For other uses you might want to modify the script or roll your own

Requirements

• tesseract OCR: sudo apt install tesseract-ocr

• tesseract language pack: sudo apt-get install tesseract-ocr-eng (for receipt_parser.py, if you're rolling your own implementation in a different language, you will need the language pack in that language)

• OpenCV 2 (in pipfile)

• Rest of requirements can be installed from pipfile using pipenv install

Structure

There are 4 basic modules making up this software:

• preprocessing: contains Processor abstract class and a bunch of implementing classes, each one implements the process(img) method which performs some sort of pre-processing. The available ones currently are: Denoiser, ArtifactRemover,Deskewer,Dilater,Eroder.

• parsing: contains the Parser abstract class, and ReceiptParser implementation which performs rule-based parsing on the given pre-processed image, i.e. it picks the highest currency formated value becomes the total. This class has accepts a tesseract language code, a price regex and a date regex - the regexes have to capture certain parts of the price and date as detailed in the code (defaults work with UK receipts).

• output: contains the Printer abstract class, and JsonPrinter implementation of it, these as the name imply simply output the parsed receipts

• classes: contain the Item and Receipt classes which are parsing targets.

Usage

Single receipt:

python receipt_parser.py path-to-img.png path-to-output.json

Multiple receipts:

python receipt_parser.py -b path-to-img-dir path-to-output-dir

Calibration (analytics mode), if you have a set of labelled data (expected json output files in the same format as printer):

python receipt_parser.py -a path-to-img-dir path-to-output-dir path-to-training-labels path-to-analytics-output-dir

Example

input:

Output:

{
    "day": "1",
    "month": "12",
    "year": "20",
    "total_whole_part": "13",
    "total_fractional_part": "44",
    "items": [
        {
            "name": "eggs",
            "price_whole_part": "1",
            "price_fractional_part": "50"
        },
        {
            "name": "vegetables ~",
            "price_whole_part": "1",
            "price_fractional_part": "50"
        },
        {
            "name": "bbq pizza",
            "price_whole_part": "3",
            "price_fractional_part": "50"
        },
        {
            "name": "frui}. juice x",
            "price_whole_part": "0",
            "price_fractional_part": "69"
        },
        {
            "name": "- pizza .",
            "price_whole_part": "2",
            "price_fractional_part": "75"
        },
        {
            "name": "pizza",
            "price_whole_part": "3",
            "price_fractional_part": "50"
        }
    ]
}