How to remove noise?

I have uploaded images:

http://i.imgur.com/jlGZw.jpg

http://imgur.com/bY0oU

Please tell how to remove the extra white regions, other than the text. I need only the text. Please help.

Answer 1

[Image Analyst]

For what's it's worth, here is 10 minutes worth of a very simplistic attempt to isolate letters. It obviously misses some of the letters for the reasons I mentioned in my comments above. That could possibly be fixed to get some of the missing ones, at the expense of altering the shapes of the ones it got correctly on the first pass, or maybe you can avoid even that with an even more sophisticated algorithm.

clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
imtool close all;  % Close all imtool figures.
clear;  % Erase all existing variables.
workspace;  % Make sure the workspace panel is showing.
fontSize = 20;

% Read in demo image.
folder = 'C:\Users\Kash\Documents\Temporary';
baseFileName = 'jlGZw.jpg';
% Get the full filename, with path prepended.
fullFileName = fullfile(folder, baseFileName);
% Check if file exists.
if ~exist(fullFileName, 'file')
  % File doesn't exist -- didn't find it there.  Check the search path for it.
  fullFileName = baseFileName; % No path this time.
  if ~exist(fullFileName, 'file')
    % Still didn't find it.  Alert user.
    errorMessage = sprintf('Error: %s does not exist in the search path folders.', fullFileName);
    uiwait(warndlg(errorMessage));
    return;
  end
end
grayImage = imread(fullFileName);
% Get the dimensions of the image.  
% numberOfColorBands should be = 1.
[rows columns numberOfColorBands] = size(grayImage)
% Display the original gray scale image.
subplot(2, 3, 1);
imshow(grayImage, []);
title('Original Color Image', 'FontSize', fontSize);
% Enlarge figure to full screen.
set(gcf, 'units','normalized','outerposition',[0 0 1 1]);
% Give a name to the title bar.
set(gcf,'name','Demo by ImageAnalyst','numbertitle','off') 

% Convert to gray scale image and threshold to get binary image.
binaryImage = rgb2gray(grayImage) > 100;
% Display the image.
subplot(2, 3, 2);
imshow(binaryImage, []);
title('Binary Image', 'FontSize', fontSize);

% Invert it.
binaryImage2 = ~binaryImage;
% Display the image.
subplot(2, 3, 3);
imshow(binaryImage2, []);
title('Inverted Binary Image', 'FontSize', fontSize);

% Border kill.
binaryImage3 = imclearborder(binaryImage2);
% Get rid of blobs smaller than 15.
binaryImage3 = bwareaopen(binaryImage3, 15);
% Display the image.
subplot(2, 3, 4);
imshow(binaryImage3, []);
title('Binary Image', 'FontSize', fontSize);

% Crop to bounding box.
verticalProfile = any(binaryImage3, 2);
horizontalProfile = any(binaryImage3, 1);
x1 = find(horizontalProfile, 1, 'first');
x2 = find(horizontalProfile, 1, 'last');
y1 = find(verticalProfile, 1, 'first');
y2 = find(verticalProfile, 1, 'last');
binaryImage4 = imcrop(binaryImage3, [x1, y1, x2-x1+1, y2-y1+1]);
% Display the image.
subplot(2, 3, 5);
imshow(binaryImage4, []);
title('Binary Image', 'FontSize', fontSize);

% Label each blob with 8-connectivity, so we can make measurements of it
[labeledImage numberOfBlobs] = bwlabel(binaryImage4, 8);
% Apply a variety of pseudo-colors to the regions.
coloredLabelsImage = label2rgb (labeledImage, 'hsv', 'k', 'shuffle'); 
% Display the pseudo-colored image.
imshow(coloredLabelsImage);
% Get all the blob properties.
blobMeasurements = regionprops(labeledImage, 'all');   
allAreas = [blobMeasurements.Area]
title('Labeled Letters', 'FontSize', fontSize);

[kash]

Thanks but some letters are missing and i did not get bounding box for each line for those text,i need two bounding boxes for each line(because it has two lines of text), please help

[Image Analyst]

Yes, like I said. I can't continue/complete this for you - this is where you take over.

[Hetal Jariwala]

Thanks for the information. works like charm

[shafaq nisar]

it is helpful

Answer 2

[Walter Roberson]

Everything in your images appears to be text. English in one patch, something Arabic-like in another patch, and some alphabet I do not recognize in a third patch.

Everything is text is some alphabet.

[kash]

http://imgur.com/77Dxh

thi sis original image ,for text detection using connected component i got that output

else please tell hoe to get text from

http://i.imgur.com/jlGZw.jpg

[Walter Roberson]

http://www.mathworks.com/matlabcentral/answers/34679-please-can-anyone-help-i-require-a-matlab-code-for-removing-text-from-an-image

[kash]

Walter i did not get any required answer from that link

[Walter Roberson]

All three areas in your picture are text, so none of them can be removed.

Wasps read the brick walls of my house every year and find the "enter here" signs easily. I don't know the alphabet the wasps are using, but then I don't know the Arabic-like alphabet in your pictures either.

[kash]

ok walter plz tell how to draw bounding box for each word in that image

[Walter Roberson]

No, I do not know how to find word boundaries for the Arabic-like string, or for wasp-ish, or for Dalek.

[kash]

Walter i need for only hindi and english words

Answer 3

[Geoff]

As Walter says, everything is text. Are you planning to use character recognition on this?

In your case, I'm going to assume you want to do the filtering only on the supplied images.

I would try something like this:

1. Detect blobs of connected pixels.

2. Go through your blob list and discard any that don't meet some required criteria (width, height, width/height ratio, number of pixels).

3. You are left with a list of all blobs that meet your 'goodness' criteria, and you construct an image using the pixels of those blobs.

An easy way to detect blobs is by the Union-Find algorithm: http://en.wikipedia.org/wiki/Disjoint-set_data_structure

There might be a nicer description somewhere else with pretty diagrams... Use Google.

You move through your image and do a Union on the current pixel with:

* the pixel immediately right (x+1, y)
* the pixel immediately down (x, y+1)
* the pixel immediately down and right (x+1,y+1)
* the pixel immediately down and left (x-1, y+1)

I haven't done this for years, and my brain doesn't feel like digging it up right now. But that's somewhere to start.

[kash]

Geoff can u tell how to draw bounding box for each line in that image

[Geoff]

What line? Which image? You gave us two images. You mean the real messy one? You mean the lines above and below "Hydraulics Laboratory"?

[Image Analyst]

Yeah, it's possible (for this image) that color segmentation might produce a better binary image than whatever kash did. But I foresee a great deal of trouble with the mix of languages, like you said. But I'm pretty sure all the OCR apps in the File Exchange are fairly unsophisticated, and not robust, or dependent on characters that match very specific templates, or have other limitations. Robust OCR is not an easy task. I did write some simple code to get the letters based on the (admittedly poor) binary image he supplied here, and it's in my Answer in this thread.

[kash]

GEof a bounding box for a word "Hydraulics Laboratory" and for a line which in in hindi,so i will get two bounding boxes for two lines

[Geoff]

Just to be clear here, are you intending to use this method on other images? Any useful method that you can apply to this image may well not work at all on other images. If you just want to do this image, is there a reason you need to "detect" the lines of text? Cos I know what I'd do... Click, click, click, click, click, click, click, click. Eight user-selected co-ordinates = two bounding polygons. Anyway, this thread seems to have gone beyond 'help with MatLab', and into 'please write my algorithm'-territory.

[kash]

Geof whats the solution for this

Answer 4

[Image Analyst]

Well I didn't see any white text whatsoever. Only some black text interior to some white surround. (You could make it white though with inverting, border-clearing, hole-filling, and some other tricks though.) So my output image would be all zero. I think this task, doing OCR on text with very strange fonts, graphics, noise and all kinds of other garbage in the image is beyond kash's abilities or even mine. There are whole OCR companies with dozens of people who have worked on this for decades and even their accuracy would be low for these kinds of images.

[Walter Roberson]

I think kash's original image would be the easiest one to work with, http://imgur.com/77Dxh .

Isolate the blue box, convert blue to black, do the ocr on the white.

[Geoff]

Oh! I never saw that image. I concur... Do the hard stuff BEFORE filtering out all the useful data from your image.

[kash]

S walter you are correcr cut i dont need ocr,i only need bounding box for each line of text

[Image Analyst]

One way to do it is in my code that has the comment "% Crop to bounding box."