Reading seven segment display with OCR command

I am trying to read the digits off of Images of a seven segment display, like the one below.

b = imread('frame1.jpg'); %image above
imshow(b)
BW1 = imbinarize(rgb2gray(b),0.2);
BW2 = imcomplement(BW1);
figure; imshow(BW2); roi = round(getPosition(imrect));
imshowpair(BW1,BW2,'montage')
ocrResults = ocr(BW2,roi,'CharacterSet','.0123456789','TextLayout','word');
recognizedText = ocrResults.Text;
text(600, 150, recognizedText, 'BackgroundColor', [1 1 1]); 

However the ocr command cannot read the digits correclty, as can be seen in the textbox on the processed image below.

Are there any pre-processing steps I should additionaly take?

Do I need to use the ocr command in a different way?

Thanks in advance for any suggestions

[Brett Shoelson]

Attempts to do this using image morphology are almost certain to fail under certain conditions. Better: download and install a (freely available) 7-segment font, then use the OCR Trainer App to train the character recognition on samples of those digits. You'll get better results!

Brett

Answer 1

[KALYAN ACHARJYA]

One Suggestion: connect the disconneted pixels, then apply the OCR

You can do that in multiple ways, imdilate , choose proper structuring element.

Let me know the result, is it working?

[Kevin Ammann]

Thank you! Connecting the digits internaly did the trick.

I used imerode in the following way.

...
BW2 = imcomplement(BW1);
SE1 = strel('line',4.5,0);
SE2 = strel('line',4.5,90);
BW2 = imerode(BW2,SE1);
BW2 = imerode(BW2,SE2);
figure; imshow(BW2); roi = round(getPosition(imrect));
...  

This solves the problem as stated in my question. However the code is not entirely robust. The initial image above is a frame of a video.

When I apply it to other frames, wrong recognitions still occur. For example with the following initial image.

The adjusted code reads the following.

Do you have any suggestions on how to make the code more robust?