Find the index of pixels in an image statisfies one condition - matlab

Question

I have an image include 4 values {3,-3,1,-1} as figure
Let call the index of pixel that its values equals 1 or -1 pixel in contour. These pixels will create a contour that surrounds the yellow color (-3). Now, I want to find all index pixels in the contour and plus padding position inward and outward contour. As the red color, padding is set 1, hence, the index of these pixels include pixel in the contour {1,-1} and padding index as the red color. In that task, I want to find all pixel indices. How to implement that idea in matlab code. This is my code to find the index in the contour
%% Let define the image I
idx=find(I==1|I==-1);
padding=1;
%%Continue
Update: My expected result as the above figure in white region. Hence, the indices are such as 13,14,15,..21,24,...
UPDATE
Firstly, thank Andrew and Rayryeng for your answer. I would like to extedn my problem. As the above description, the contour is created by {1,-1}. Now, I want to ignore 1 and -1, so the image I only has {3,-3}. And I defined the contour is pixel in the edge of {3,-3} such as figure. Keep the same idea of padding and pixel index. How to find the indices of pixels in contour and near contour (call narrow band of contour)(expected result is white color)

Answer 1

Not too difficult you are on the right track. If you have the image processing toolbox, I recommend taking a look at morphological operators. Specifically you want to use imdilate my code has all the details you need.
%rather than using find, we create a binary mask. Its not the indicies of
%the matching elements as find gives. its is 1/true if the value matches the
%criteria, and 0/false otherwise.
mask = (im=1 | im=-1);
%create a 3x3 rectangle structuring element. We use a 3x3 because we want
%to expand the image by one pixel. basically the structring element (Strel)
%is our kernal, if you know image processing this is the same thing.
%a = [0 0 0 0;
% 0 1 1 1;
% 0 1 1 1;
% 0 1 1 1];
%our kernal is center at 2,2 (for this example) which are these elements
% 0 0 0 of a think a(1:3,1:3) now what the dialate operation
% 0 1 1 says is, if the majority of these pixels are ones... they
% 0 1 1 should probabaly all be ones so all those 0s will become ones
%the size of the kernal 3x3 ensures we are only growing our image one
%pixel, hope that makes sense
se = strel('square',3);
%now we dilate, or 'expand' our mask with our structuring element
expanded_mask = imdilate(mask,se);
%if you still want the indicies you can use find on our expanded mask
idx = find(expanded_mask==1);
EDIT: without morphological operations/image processing toolbox
This method uses lots of for loops, so it isn't the fastest, and doens't do error checking, but it will work. My dilate function says if the majority of the pixels are ones make them all ones.
function expanded_mask=DilateBinaryImage(bin_im, kernal_size)
[max_row,max_col] = size(bin_im);
%since we are opening the mask (only adding 1s), we can start off with the
%same values of the mask, and simply add extra 1's as needed
expanded_mask = bin_im;
%we don't want to go off the edge of our image with this kernal
%so we offset it a bit
kern_padding = floor(kernal_size/2);
%this ignores the edges
for (curr_row=kern_padding+1:1:max_row - kern_padding)
for (curr_col=kern_padding+1:1:max_col - kern_padding)
%we do 2 sums, one for rows, one for columns
num_ones = sum(sum(bin_im(curr_row-kern_padding:curr_row+kern_padding,curr_col-kern_padding:curr_col+kern_padding)));
%if the majority of vlaues are 1, we use floor to help with corner
%cases
if (num_ones >= floor((kernal_size*kernal_size)/2))
%make all the values one
expanded_mask(curr_row-kern_padding:curr_row+kern_padding,curr_col-kern_padding:curr_col+kern_padding) = 1;
end
end
end
end
and then called it like this
kernal_size= 3;
mask = (I==1 | I==-1);
expanded_mask = DilateBinaryImage(mask, kernal_size);
idx = find(expanded_mask==1);
my dilate function doesn't work at the edges of the binary image. it just copies them exactly.

Answer 2

Lets say your image is N-by-M pixles. In MATLAB arrays are stored in column order (see http://www.mathworks.com/help/matlab/math/matrix-indexing.html for more information). You can use I in column format as follows. First you contour pixels are given by
idx=find(I(:)==1|I(:)==-1);
Now, if you wish to pad downward and upward it is quite simple:
idx_up=idx - padding;
idx_up = idx_up(idx_up>0);
idx_down=idx + padding;
idx_down = idx_down(idx_down<=N*M);
Note that idx_up and idx_down will also contain contour pixels.
Similarly you can pad to the left\right:
idx_left=idx - padding*N;
idx_left = idx_left(idx_left>0);
idx_right=idx + padding*N;
idx_right = idx_right(idx_right<=N*M);
And combine the overall pixels:
PaddedContour = false(N,M);
PaddedContour(unique([idx;idx_up;idx_down;idx_left;idx_right])) = true;