How to get only text using OCR recognition feature of Microsoft Cognitive Services - Vision API?

Question

I am using sample provided at Computer Vision API C# Quick Start I am able to get JSON result as shown in Sample but unable to get only text content.

Sample format of JSON is as below:

{
  "textAngle": 0.020943951023932542,
  "orientation": "NotDetected",
  "language": "de",
  "regions": [
    {
      "boundingBox": "46,54,59,71",
      "lines": [
        {
          "boundingBox": "48,54,49,19",
          "words": [
            {
              "boundingBox": "48,54,49,19",
              "text": "Hello"
            }
          ]
        },
        {
          "boundingBox": "46,106,59,19",
          "words": [
            {
              "boundingBox": "46,106,59,19",
              "text": "World"
            }
          ]
        }
      ]
    }
  ]
}

For now, I am using JSON converter to extract text nodes by adding newline for each word using below class structure.

public class Region
{
    public string BoundingBox { get; set; }
    public List<Line> Lines { get; set; }
}

public class Line
{
    public string BoundingBox { get; set; }
    public List<Word> Words { get; set; }
}

public class Word
{
    public string BoundingBox { get; set; }
    public string Text { get; set; }
}

Is there any request parameter provided in API to get direct text in the response itself?

Not accoring to the [API documentation](https://westcentralus.dev.cognitive.microsoft.com/docs/services/56f91f2d778daf23d8ec6739/operations/56f91f2e778daf14a499e1fc) — stuartd, Feb 16 '18 at 01:42
You can use this class https://github.com/Microsoft/Cognitive-Vision-DotNetCore/blob/master/Samples/OcrSample/Program.cs as alternative, not sure about API — Chirag Rupani, Feb 16 '18 at 01:53

score 8 · Accepted Answer · answered Feb 16 '18 at 07:07

If you want C# types for the returned response, you can use the official client SDK in github. It's also available in NuGet.

Once you have the OcrResults, and you just want the text, you could write some hacky C# code with Linq like this:

string OcrResultsToString(OcrResult result)
{
    return string.Join("\n",
        result.Regions.ToList().Select(region =>
            string.Join(" ", region.Lines.ToList().Select(line =>
                 string.Join(" ", line.Words.ToList().Select(word =>
                     word.Text).ToArray())).ToArray())).ToArray());
}

Or, if that hurts your eyes, you could use conventional loops like this:

 string OcrResultsToString(OcrResults results)
 {
    StringBuilder stringBuilder = new StringBuilder();

    if (results != null && results.Regions != null)
    {
        foreach (var item in results.Regions)
        {
            foreach (var line in item.Lines)
            {
                foreach (var word in line.Words)
                {
                    stringBuilder.Append(word.Text);
                    stringBuilder.Append(" ");
                }
                stringBuilder.AppendLine();
            }
            stringBuilder.AppendLine();
        }
    }
    return stringBuilder.ToString();
}

Thanks, would like to see this to be added in API itself – userda Feb 19 '18 at 16:13 — userda, Feb 19 '18 at 16:13
Totally agree .... Can't believe it is not – Mann Wong Apr 06 '22 at 18:42 — Mann Wong, Apr 06 '22 at 18:42

How to get only text using OCR recognition feature of Microsoft Cognitive Services - Vision API?

1 Answers1