.NET OCRing an Image

Question

I'm trying to use MODI to OCR a window's program. It works fine for screenshots I grab programmatically using win32 interop like this:

public string SaveScreenShotToFile()
{
    RECT rc;
    GetWindowRect(_hWnd, out rc);

    int width = rc.right - rc.left;
    int height = rc.bottom - rc.top;

    Bitmap bmp = new Bitmap(width, height);
    Graphics gfxBmp = Graphics.FromImage(bmp);
    IntPtr hdcBitmap = gfxBmp.GetHdc();

    PrintWindow(_hWnd, hdcBitmap, 0);

    gfxBmp.ReleaseHdc(hdcBitmap);
    gfxBmp.Dispose();

    string fileName = @"c:\temp\screenshots\" + Guid.NewGuid().ToString() + ".bmp";
    bmp.Save(fileName);
    return fileName;
}

This image is then saved to a file and ran through MODI like this:

    private string GetTextFromImage(string fileName)
    {

        MODI.Document doc = new MODI.DocumentClass();
        doc.Create(fileName);
        doc.OCR(MODI.MiLANGUAGES.miLANG_ENGLISH, true, true);
        MODI.Image img = (MODI.Image)doc.Images[0];
        MODI.Layout layout = img.Layout;

        StringBuilder sb = new StringBuilder();
        for (int i = 0; i < layout.Words.Count; i++)
        {
            MODI.Word word = (MODI.Word)layout.Words[i];
            sb.Append(word.Text);
            sb.Append(" ");
        }

        if (sb.Length > 1)
            sb.Length--;

        return sb.ToString();
    }

This part works fine, however, I don't want to OCR the entire screenshot, just portions of it. I try cropping the image programmatically like this:

    private string SaveToCroppedImage(Bitmap original)
    {
        Bitmap result = original.Clone(new Rectangle(0, 0, 250, 250), original.PixelFormat);
        var fileName = "c:\\" + Guid.NewGuid().ToString() + ".bmp";
        result.Save(fileName, original.RawFormat);

        return fileName;
    }

and then OCRing this smaller image, however MODI throws an exception; 'OCR running error', the error code is -959967087.

Why can MODI handle the original bitmap but not the smaller version taken from it?

Answer 1

Looks as though the answer is in giving MODI a bigger canvas. I was also trying to take a screenshot of a control and OCR it and ran into the same problem. In the end I took the image of the control, copied the image into a larger bitmap and OCRed the larger bitmap.

Another issue I found was that you must have a proper extension for your image file. In other words, .tmp doesn't cut it.

I kept the work of creating a larger source inside my OCR method, which looks something like this (I deal directly with Image objects):

public static string ExtractText(this Image image)
{
    var tmpFile = Path.GetTempFileName();
    string text;
    try
    {
        var bmp = new Bitmap(Math.Max(image.Width, 1024), Math.Max(image.Height, 768));
        var gfxResize = Graphics.FromImage(bmp);
        gfxResize.DrawImage(image, new Rectangle(0, 0, image.Width, image.Height));
        bmp.Save(tmpFile + ".bmp", ImageFormat.Bmp);
        var doc = new MODI.Document();
        doc.Create(tmpFile + ".bmp");
        doc.OCR(MODI.MiLANGUAGES.miLANG_ENGLISH, true, true);
        var img = (MODI.Image)doc.Images[0];
        var layout = img.Layout;
        text = layout.Text;
    }
    finally
    {
        File.Delete(tmpFile);
        File.Delete(tmpFile + ".bmp");
    }

    return text;
}

I'm not sure exactly what the minimum size is, but it appears as though 1024 x 768 does the trick.

Answer 2

yes the posts in this thread helped me gettin it to work, here what i have to add:

was trying to download images ( small ones ) then ocr...

-when processing images, it seems that theyr size must be power of 2 ! ( was able to ocr images: 512x512 , 128x128, 256x64 .. other sizes mostly failed ( like 1103x334 ))

• transparent background also made troubles. I got the best results when creating a new tif with powerof2 boundary, white background, paste the downloaded image into it, save.

• scaling the image did not succeed for me, since OCR is getting wrong results , specially for "german" characters like "ü"

• in the end i also used: doc.OCR(MODI.MiLANGUAGES.miLANG_ENGLISH, false, false);

• using modi from office 2003

greetings

womd