ReceiptLedger is an ongoing personal project that uses OCR to extract information from receipts, with the goal of storing that data in a database for bookkeeping.
- Scan receipts with a document scanner, regardless of their original size or format.
- Save each scanned receipt as a single PDF on a NAS.
- Extract text using Tesseract OCR with Dutch language data.
- Turn the extracted information into structured receipt data for database storage.
Work in progress — the OCR foundation is working.
- Working Tesseract OCR setup with Dutch (
nld) language data. - Required underlying dependencies installed and working.
- A
pdf2cvstool to test the OCR workflow. - Receipt classification by store or receipt type.
- Matching low-confidence OCR results against known articles in the database.
- A GUI may be considered only if the CLI no longer provides a workable way for the user to correct OCR errors.
My interest in OCR started about twenty years ago, when I encountered OCR engine code as part of a document management system (DMS). Later, I discovered OpenCV for image processing.
Projects such as flomonster/ocr, which explore image enhancement and neural network training, provide an interesting point of comparison.
For ReceiptLedger, I chose Tesseract as the OCR engine. For background on its architecture, see An Overview of the Tesseract OCR Engine.