Calling Google cloud Vision API's on numpy matrices

Question

I'm using the Google Text detection API for performing OCR on images.

I've found that my OCR results are much better when I perform some pre-processing on the images using opencv.

My question is - how can I call the Google cloud Vision API's on images I have in memory as Numpy arrays? The official Google docs only show the vision api accepting an image in disk as the input.

I want to avoid unnecessary disk writes.

Answer 1

One way of providing the image to Google Cloud Vison API is as a base64 encoded string. The quickstart example uses the Python client API, which accepts a stream of bytes for the image content.

Instead of writing to file you can encode your OpenCV image img to a byte string you can directly pass to Cloud Vision API:

image = types.Image(content=cv2.imencode('.jpg', img)[1].tostring())