Question

This is the original image.

I changed it into 1) grayscale and apply 2) threshold.

As seen in the original image, there are some shadow that still exist after apply two method above.

But most of the image are perfect after 2 method.

I need to extract the text, so I need to get rid of the noise. I almost finish the work but the problem is some cases have a black border and I wanted to replace that into white color.

And I insist that I want just only border to turn into white

I think of create some white rectangle and fill the border with those rectangle but I don’t know how to do it.

How can I achieve that using Matlab?
Any other method would be appreciate too!

Answer 1

If you can be sure only the borders will be black, why not simply crop the image until all isolated shapes are recognizable characters? Something along the lines of

done = false;
ii = 1;
while (~done)

    % fill the outer border
    img(:,ii)       = 255;    img(ii,:)       = 255;
    img(:,end-ii+1) = 255;    img(end-ii+1,:) = 255;

    % (run your algorithms here. It positive match, done = true)

end

That could be computationally intensive, since you have to do pattern recognition on each iteration, but you indicated it only occurs in “some cases”.

Otherwise, I suspect some morphological operation can also be used, probably erosion or thinning or similar. But that has the drawback of altering the characters you want to match. But, if all images you have to process look like the one you show, I hardly suspect that’ll give you any problems.

Some ways to detecting straight lines are mentioned in this question. I’d say you could detect all lines, and remove those (with a small tolerance around it) that are perfectly horizontal/vertical and on one of the edges.