The quality of a receipt directly impact the quality of the analysis. Here is a guideline to help you send great receipts to have good results.
General conditions
- 🏙 Background should be contrasted (ideally dark), without pattern
- ☀️ Sufficient luminosity and contrast (avoid black and white, filtered and binarized image)
- 👓 No blur.
- 📷 Enough image resolution (we must be able to zoom and sufficiently read characters on the receipt).
- 🖍 Avoid editing image (no pre-cropping, no metadata destruction like exif loss)
Position of receipt on the picture
- Picture should be taken vertically with the normal reading orientation of the receipt (top of the receipt = top of the picture).
- Receipt should represent a rectangle in the picture as much as possible. Avoid taking picture in perspective. So be as much as possible in front of the receipt when taking the picture.
- Receipt should be flat (rectangle!), no folds.
- Receipt should present the integrality of the necessary information: Avoid partial pictures
We cannot analyze what is not on the picture.
Requirements on the receipt's quality.
- Receipt should be readable by human. No faded characters.
- Receipt should not be wrinkled or damaged in any other way that affects the quality of the text.
- Receipt should not have any added writings: no underlinings, circles, etc.
- Receipt should not be folded or hidden in any part
- Receipt image should be as original as possible
Recommandations
- We use all the image to extract data, even metadata, we can extract GPS informations and more from exif metadata, so be sure to preserve those data when sending image to our system to get the maximum performance
Pictures examples
✅ Good pictures examples:
Good:
🚫 Bad pictures examples:
Bad 1
- Wrong orientation
- Extra writings
- Cut receipt (not all informations can be found).
Bad 2
- Not unified background
- Receipt is not rectangular: wrinkled, not flat.
Comments
0 comments
Please sign in to leave a comment.