Computer Vision API - Can't convert to binary from Base64

(0) Share

Report

Posted on by Stanza

Hi,

I am using Computer Vision API action Optical Character Recognition (OCR) to Text

I have a canvas app that takes a picture and triggers a flow.

When the flow runs it does not recognise the text.

I have built my Flow based on the examples online.

As per my understanding I need to convert the base64 text to binary for the OCR to read the image.

However I get an error when I try to convert it :

Unable to process template language expressions in action 'Compose_2' inputs at line '1' and column '8993': 'The template language function 'base64ToBinary' was invoked with a parameter that is not valid. The value cannot be decoded from base64 representation.'.

My flow:

If I do not convert the image

Any ideas ?

Stanley

Categories:

Building flows

I have the same question (0)

All responses (3)

Answers (1)

Sort by

rrovira Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!

Hello, @Stanza!

What kind of data is the second compose returning? Are you sure it's returning a Base64 string?

Cheers,
Rodrigo!

Was this reply helpful? Yes No
Stanza 21 on at

Like (0)

Report
Copy link

Link copied!
Hi @rrovira,

I managed to get rid of the error. However, now the Flow runs with no error but I am getting blank text, so the API is not detecting my image.

PowerApps setup is:
PowerApps button to capture image:
OnSelect:

ClearCollect(PicCollection, Camera1.Stream)

PowerApps button to send to Flow:
OnSelect:

TakePictureV2.Run(First(PicCollection).Url)

Flow setup:

I am under the impression that we need to convert using dataUriToBinary function so I have supplied the photo.url as input from PowerApps.
I have also created the file in OneDrive as per the examples on the internet.
The files I have scanned are below. Is the API not able to recognise these images?
I have tried using a business card and same result - all return blank

Was this reply helpful? Yes No
Verified answer

Stanza 21 on at

Like (0)

Report
Copy link

Link copied!

Ok here's the news, turns out the API doesn't recognise handwritten running writing (at the moment).
I scanned an image with block text and it worked !

Was this reply helpful? Yes No