Recently found an interesting article comparing different OCR services
I also want to try on Azure Cognitive Service computer vision. I find they provide free service for you to try (but you need to find it)
- Go to https://azure.microsoft.com/en-us/try/cognitive-services/my-apis/ to register trial computer vision service. Then get your API key and endpoint
- Azure cognitive services Computer Vision OCR has changed a lot since first release, adding more features, API also changes much. Microsoft documentation recommends using Read 3.0 API (asynchronous operation), rather than previous OCR, Recognizetext API. At the bottom of the article, go to the link to try out their API
https://docs.microsoft.com/en-us/azure/cognitive-services/computer-vision/concept-recognizing-text - In this page, select testing console in the region where you created your resource in step 1.
- Input language as “en”, Ocp-Apim-Subscription-Key. In request body, input image URL. Click Send
- In response content, note Operation-Location. The last part is the operation id you will use in next step
- In testing console, click “Get Read result”. Select region, input operationId, Ocp-Apim-Subscription-Key. Click send. Then it returns OCR result
Result
For the first 3 category, I get similar quality as Google Cloud vision.
For table extraction, Azure provides Cognitive services Forms recognizer. Currently the prebuilt model only has receipt and business card (I don’t think it will support that table in the article). I tried with receipt model with steps in
It only extracts a few line and expected format is definitely different. To train a custom model, let’s save for another day.
