identifying number on dice

% I want to crop the front face of dice and identify the number on it. I tried the following

A=imread('dice.png');
figure,imshow(A); title('Original Image');
%Convert the Image to binary
B=im2bw(A);

%Fill the holes
C=imfill(B,'holes');

%Label the connected components
[Label,Total]=bwlabel(C,8);
figure,imshow(C); title('Labelled Image');
% then I did the following
D=imsubtract(B,C);
imshow(D);
%

where Background is black, dice is white with black dots. But WHEN BACKGROUND IS NOT PERFECTLY BLACK THEN IT CREATES PROBLEM. CAN YOU PLZ TELL ME HOW REMOVE BACKGROUND AND CROP ONLY DICE PART?

[Aayush Maharjan]

How can I do the same using python.??? I don't understand Matlab code.

[Image Analyst]

What do you understand? Python? If so, you should know the corresponding functions in Python. I don't know Python's library functions well enough to do the conversion for you. But presumably you know Python much better than MATLAB so you should be able to translate it yourself. If you don't, then try asking in a Python language community forum.

Answer 1

[Sven]

Hi Sameer, Here's what I would do. It's a little like IA's bwareaopen suggestion. It tries to make minimal assumptions other than:

1. Dice are white and bigger than 1000 pixels
2. (front facing) dots are surrounded by dice and bigger than 50 pixels

% Fetch the image
I = imread('http://www.mathworks.com/matlabcentral/answers/uploaded_files/13250/dice1.jpg');
G = rgb2gray(I);

% Dice are white and big
diceBlobs = bwareaopen(G>180, 1000);
cc = bwconncomp(diceBlobs);
diceLabels = labelmatrix(cc);

for i = 1:cc.NumObjects
    thisDice = diceLabels==i;
    % Dots are surrounded by dice and +50 pixels
    thisDots = bwareaopen(imfill(thisDice,'holes') & ~thisDice, 50);
    dotsCC = bwconncomp(thisDots);
    figure, imshow(thisDice*0.5 + thisDots)
    title(sprintf('Dice #%d has %d dots',i,dotsCC.NumObjects))
end

Does this code help you out? You could certainly add some further criteria such as dots needing to be round (I would use regionprops and look at, say, the Solidity property), and maybe dice should be square, but this seems reasonably robust as long as your pictures are all quite similar.

Thanks, Sven.

[Sameer]

Thank you very much

Answer 2

[Image Analyst]

You forgot to attach your image(s). It's more difficult to give advice on image processing without an image. Maybe you can do a median filter or maybe call bwareaopen, or do some size filtering. If the black spots in the background are the same size as your spots, you will have to make a mask. Threshold for bright things, fill them in, then take the largest.

[Sameer]

ooops, I'm sorry. here is the image.