Recently, I decided to put Google cloud vision's OCR to the test, only to be disappointed by the subpar results. Despite my documents being in French, the OCR seemed to struggle with detecting apostrophes and commas accurately. For instance, when using the following image as input https://i.sstatic.net/i1Ljv.jpg
The code used was as follows:
Request
.post(`https://vision.googleapis.com/v1/images:annotate?key=AIzaSyAtArxxxxxxxxxxxxxxxxxpGrKrydU4`)
.send({
requests: [{
image: { content: base64.replace('data:image/jpeg;base64,', '') },
features: [{ type: 'DOCUMENT_TEXT_DETECTION' }],
"imageContext": { "languageHints": [ "fr" ] }
}]
})
The result obtained (with errors highlighted in yellow) can be viewed here: https://i.sstatic.net/oKCyX.png
In contrast, Microsoft Azure OCR provided a flawless result without requiring me to specify the language, as seen here.
I'm curious if others have experienced similar accuracy issues with Google Cloud Vision's OCR.