OCR not reading Images Captured from Webcam

[rajasajidmanzoor]

Hello. I am trying to capture images via webcam and then read it using OCR but OCR is not reading them. I have also bought an HD 1080p anti flare and anti noise web cam but sill it is not reading Text and number. I am trying to read from following images. ![Uploading 635274052466609107.pn

g . . .]()

But it is not reading all images. I also tried using Grey scale images but still issue persists. Please tell me how can i read these images or also suggest me any image processing techniques if needed..

Sajid

[charlesw]

I got reasonable results after I (manually) cropped the image so the border isn't included and then binarized the image (using BinarizeOtsuAdaptiveThreshold, with params img.Width / 5, img.Height / 5, 10, 10, 0.1 as previously mentioned in #71).

You will of course need to find a way to crop the image automatically maybe either doing some edge detection or page analysis.

[rajasajidmanzoor]

hmm. thanks. Can you please share a code example for BinarizeOtsuAdaptiveThreshold and also please tell me where should i use BinarizeOtsuAdaptiveThreshold..

Thanks

[charlesw]

Something like this:

using (var engine = new TesseractEngine(@"./tessdata", "eng", EngineMode.Default)) {
    using (var img = Pix.LoadFromFile(testImagePath)) {
        using (var binarizedImg = img.BinarizeOtsuAdaptiveThreshold(img.Width / 5, img.Height / 5, 10, 10, 0.1F)) {
            binarizedImg.Save("Binarized-test.png", ImageFormat.Png);

            using (var page = engine.Process(binarizedImg)) {
                var text = page.GetText();

                // whatever

            }
        }
    }
}

Where/How to use this is really going to depend on how you've written your application and what other processing you plan on doing.

[charlesw]

Note that the testImagePath must be a path to a 8 BPP (bit's per pixel) image. Binarization WILL fail if it's anything else.

[rajasajidmanzoor]

ok. How can i check that image in 8 bits per pixel image in c# and if it is not a 8 BPP image than how can i make it 8BPP

thanks

[charlesw]

Your after the Depth property which will need to equal 8 in this case. As for converting a 32-bit image (or other source) you'll probably want to use a Quantization method unfortunately I haven't wrapped these methods from Leptonica at this time. I've added issue #75 however can't say when I'll get to this. Another option may be to capture the images as 8-bit images if possible.

[rajasajidmanzoor]

hello Can you please describe what is Quantization method. I tried to convert an image to Bit using following method. http://codingresolved.com/discussion/2185/save-png-image-as-8bit-format-in-c and then applied binarization as you told me above. but it is not reading text from image. the image that i have used is (before converting to 32 bit see below image) before converting to 8 bit it only reads last line.

See below image after conversion to 8 bit. Not a single word has been read after converting to 8 bit.

[charlesw]

Did you also crop it, this is the image I ended up processing that worked:

[rajasajidmanzoor]

Hello Please confirm how did you achieved this result. i performed following trick to get the second image http://codingresolved.com/discussion/2185/save-png-image-as-8bit-format-in-c

[charlesw]

I just opened it in my favorite picture editor (Paint.Net in this case), cropped the image and saved it as a png file with 256 colours (8 bpp) and then fed that to the previously mentioned routine.

[rajasajidmanzoor]

hmm. Ok thanks But if possible can you please share a code example for pixOctreeColorQuant so that i get a start. I know that i am disturbing you and taking a lot of your time but i don't have any other way to sort out. I will be thankful to you if you help me.

[charlesw]

OK, given you time limits here's what I would do:

1. Forget about BinarizeOtsuAdaptiveThreshold for now (this removes the 8 BPP requirement)
2. Run an analyse over the image, hopefully it identifies the characters on the plate.
3. Iterate over the result and build a bounding box that contains the center characters (see sample).
4. Crop image to bounding box.
5. OCR the cropped image

If step 2 doesn't come up with any results you could try a fallback of just cropping the edges and checking the results however this is really unreliable but might be better than nothing.

[rajasajidmanzoor]

ok. but there where is sample. Are you talking about the samples included in Wrapper..

[charlesw]

Yes, specifically the console one.

On Wed, Feb 12, 2014 at 8:44 PM, sajiddesigner notifications@github.comwrote:

ok. but there where is sample. Are you talking about the samples included in Wrapper..

Reply to this email directly or view it on GitHubhttps://github.com/charlesw/tesseract/issues/74#issuecomment-34852981 .

[rajasajidmanzoor]

and also please confirm that which file format is best for optimum results in tesseract. JPG, PNG, TIFF or Gif.