How to ONLY detect a particular language using cloud vision api in python and ignore others?

user_6737387 0
Viewed 393

I want to detect korean language from the image and only korean. I don't want to detect english even if it occurs in the image. I can't seem to find any way on internet that could do this for me.

What I've tried:

I'm aware of the the image_context attribute and am already using it to mention only korean but that didn't ignore english text. I did:

response = client.document_text_detection(image=image, image_context={"language_hints": ["ko"]})

What I want?

Is there a way to not detect English language from the image?

Or is there a way to block certain symbols from being detected ?

google-cloud-vision computer-vision ocr python
1 Answers

I believe the image_context attribute is not used to restrict the language that you want to extract on a single image. It does the opposite as it helps to amplify and get better results for understanding the language. 1

Providing a language hint to the service is not required, but can be done if the service is having trouble detecting the language used in your image.

If the language hint is left blank, we will attempt to auto-detect the most appropriate language. The TEXT_DETECTION endpoint will auto-detect only a subset of supported languages, while the DOCUMENT_TEXT_DETECTION endpoint will auto-detect the full set of supported languages.

One way to achieve your use-case might be by manipulating the return response from the Vision API. You can extract only the Korean characters or exclude the results with English text.

"textAnnotations": [
    {
      "locale": "ko",
      "description": "牛牛面馆\n",
      "boundingPoly": {
        "vertices": [
          {
            "x": 159,
            "y": 212
          },
          {
            "x": 947,
            "y": 212
          },
          {
            "x": 947,
            "y": 354
          },
          {
            "x": 159,
            "y": 354
          }
        ]
      }
    },

Or you can file a feature request for this.

user_143523560
Related