Does Google Cloud Vision OCR API have better accuracy and performance than Tesseract OCR API

nik picture nik · Aug 8, 2017 · Viewed 15.3k times · Source

I have integrated Google Cloud Vision API in my java application for text recognition from complex formatted documents. One of my colleague suggested to use "Tesseract API".Can anyone please give difference between these two API's.And which is better in terms of accuracy or have any advantage over other.TIA

Answer

Pratap Singh Mehra picture Pratap Singh Mehra · May 13, 2020

I have used both of them. Google Vision is much faster than Tesseract and If it was a year back then the accuracy was also better.

Tesseract lately adapted LSTM with preferred language choice and trained data which when optimized could get faster about 2X or more. But surely it can not get faster than Google vision unless they start supporting CUDA.

Regarding accuracy I have tested both of them with best and worst document images and were Google Vision landed on 66.6% accuracy, Tesseract went to 82% which is high enough for me.

If you prefer accuracy Tesseract is a winner and if you prefer time Google Vision is the best option.

Also there are couple of other CUDA supported projects which may be better than them.