Sectors: Finance • Information Technoloy • Government • Healthcare • Industries • Education • (show all)
Since 1970s, optical character recognition has been used in more and more areas. Thanks to the ubiquitous presence of smartphones, receipts can be easily snapped and uploaded. To ensure a decent user experience, we need to recognize data on receipts accurately in a timely manner. Due to the low print quality of POS thermal printers, receipt OCR is a challenging task. Some vendors claim their solutions are "world's most advanced", "the most robust" and "the best" receipt OCR. How do they perform in the real world? In this article, we'll review the accuracy of the receipt recognition accross different providers: Abbyy vs Rossum vs Rydoo vs Taggun vs Asprise. Other providers are compared in another article: Compare receipt OCR: Asprise vs Nanonets vs Veryfi vs Tabscanner.
Two images are used in this test (click on the thumbnails to zoom):
The above are two original actual image files used (downloads: r1.jpg | r2.jpg). We'll refer the Walmart receipt on the first image as receipt #1 and the two receipts on the second image as receipt #2A and receipt #2B respectively.
For the sake of simplicity, we only check the result of the two most important data points: merchant name and total amount.
Founded in Moscow, Russia, Abbyy offers many OCR products. One of the products is called Cloud OCR SDK.
| Receipt #1 | Receipt #2A | Receipt #2B | ||||
|---|---|---|---|---|---|---|
| Merchant name | Total amount | Merchant name | Total amount | Merchant name | Total amount | |
| Abbyy | OK | OK | Failed to detect | OK | Low accuracy | Failed to detect |
Observations: Abbyy successfully detects receipt #1. It fails in detecting the presence of two receipts on the second image. Additionally, the recognition accuracy is not very high for the second image.
Test output files: Output of image #2: aby-r2.xml
Please refer the below screenshots for more details. (Test on: 2021-07-16)
Rossum uses artificial intelligence to understand complex structured documents and to enable companies to capture data from financial documents.
| Receipt #1 | Receipt #2A | Receipt #2B | ||||
|---|---|---|---|---|---|---|
| Merchant name | Total amount | Merchant name | Total amount | Merchant name | Total amount | |
| Rossum | Failed to detect | Failed to detect | Failed to detect | Failed to detect | OK | OK |
Observations: Rossum fails to detect both receipt #1 and receipt #2A, however it successfully detects receipt #2B.
Please refer the below screenshots for more details. (Test on: 2021-07-16)
Rydoo offers travel and expense management. In 2017, it was acquired by Sodexo, a leader in Quality of Life services.
| Receipt #1 | Receipt #2A | Receipt #2B | ||||
|---|---|---|---|---|---|---|
| Merchant name | Total amount | Merchant name | Total amount | Merchant name | Total amount | |
| Rydoo | OK | OK | OK | OK | Failed to detect | Failed to detect |
Observations: Rydoo detects receipt #1 and receipt #2A successfully, however it completely ignores receipt #2B.
Please refer the below screenshots for more details. (Test on: 2021-07-16)
Taggun is a remote company based in New Zealand with a small team of 3. It offers a receipt OCR API with accuracy of 82.26%.
| Receipt #1 | Receipt #2A | Receipt #2B | ||||
|---|---|---|---|---|---|---|
| Merchant name | Total amount | Merchant name | Total amount | Merchant name | Total amount | |
| Taggun | OK | OK | Failed to detect | OK | OK | Failed to detect |
Observations: Taggun detects receipt #1 successfully, however it confuses the result of receipts B1 and B2 by mixing the two as one.
Please refer the below screenshots for more details. (Test on: 2021-07-16)
Leveraging more than 20 years of experience, Asprise OCR offers realtime receipt OCR.
| Receipt #1 | Receipt #2A | Receipt #2B | ||||
|---|---|---|---|---|---|---|
| Merchant name | Total amount | Merchant name | Total amount | Merchant name | Total amount | |
| Asprise | OK | OK | OK | OK | OK | OK |
Observations: Asprise detects receipt #1 successfully. Asprise detects both of the two receipts on the second image.
Please refer the below screenshots for more details. (Test on: 2021-07-16)
| Receipt #1 | Receipt #2A | Receipt #2B | ||||
|---|---|---|---|---|---|---|
| Merchant name | Total amount | Merchant name | Total amount | Merchant name | Total amount | |
| Abbyy | OK | OK | Failed to detect | OK | Low accuracy | Failed to detect |
| Rossum | Failed to detect | Failed to detect | Failed to detect | Failed to detect | OK | OK |
| Rydoo | OK | OK | OK | OK | Failed to detect | Failed to detect |
| Taggun | OK | OK | Failed to detect | OK | OK | Failed to detect |
| Asprise | OK | OK | OK | OK | OK | OK |
The result above shows that receipt OCR accuracy varies across providers. Besides accuracy, you should also consider the speed (average time taken to read a receipt) and the background of the company when choosing a receipt OCR provider.
OCR your own receipts - No registration required Developer's Guide - easy integration
