I need to correct this code to fill each connected component in the image New_bw

clear all;

i = imread('resturant.jpg');

g = rgb2gray(i);

Canny_g = edge(g,'Canny');

bw = im2bw(g);

CC = bwconncomp(bw)

New_bw = zeros(size(g));

for j = 1:1:CC.NumObjects

      Temp = zeros(size(g));

      Temp(CC.PixelIdxList{j}) = 1;

      New_bw = New_bw + Temp;

      clear Temp;

end

figure, imshow(New_bw), title('image of components before'), hold on;

Regs = regionprops(CC,'BoundingBox','Centroid')

MatFill = zeros(10,1);

for i = 1:10  %CC.NumObjects

      pts = Regs(i).BoundingBox;

      plot(pts(1),pts(2),'*b');

      MatFill(i,1) = round(pts(1));

      MatFill(i,2) = round(pts(2));

end

figure, imshow(New_bw), title('image of components after'), hold on;

MatFill

ImFill = imfill(New_bw,MatFill,4);

figure, imshow(ImFill), title('after filling'), hold on;

[abdelrahim hashem]

it gives an error in the statement:: ImFill = imfill(New_bw, MatFill,4)

Answer 1

[Walter Roberson]

The "locations" parameter must be linear indices. https://www.mathworks.com/help/images/ref/imfill.html#inputarg_locations

MatFill(i) = sub2ind(size(CC, round(pts(2)), round(pts(1)) );

Note the bounding box is in X, Y order, but that indexing is in Y, X order.

I would point out, though, that you should not assume that the lower left corner of the bounding box of the object definitely is or definitely is not occupied. If it is not occupied then the imfill is going to fill around the connected component, everything over to the edge of the next connected component over.

I think you should be considering asking to fill holes instead.

[abdelrahim hashem]

Did you mean size(g), or size(CC)??? also why we start pts(2), then pts(1)??

[abdelrahim hashem]

also, the statement in your answer give the error "out of range subscript"

[Walter Roberson]

size(CC) and size(g) should be the same. Just do not use size(i) as that is a different size.

If you look at https://www.mathworks.com/help/images/ref/regionprops.html#inputarg_properties you can see that the first component of the BoundingBox property is the X coordinate, followed by the Y coordinate. In standard graphs, the X coordinate is the distance left or right, which corresponds to an array column; the coordinate is the distance up or down, which corresponds to an array row. So you cannot index an image array at (X, Y): you need to index it at (Y,X). sub2ind() takes Row, Column order, not (X, Y) order. That is why you need pts(2) followed by pts(1)

You should not really need to round() the bounding box coordinates, as they should always be integers, but I copied that part of the original code.

[abdelrahim hashem]

Thanks for you Roberson, but the CC is the structure with four fields, and g is the original image, so they different things. How they the same size??

[Walter Roberson]

You are right, it should be size(g) not size(CC)

[abdelrahim hashem]

OK, but still error in using:

for i = 1:CC.NumObjects

      pts = Regs(i).BoundingBox;

      MatFill(i) = sub2ind(size(g), round(pts(2)), round(pts(1)));

end

%%%%gives "Out of range subscript".

[Walter Roberson]

I am not sure at the moment. I have not tried to regionprops on a CC before.

I think I am going to bed now.

Answer 2

[Image Analyst]

I'm not familiar with this:

ImFill = imfill(New_bw,MatFill,4);

usually I'd do

filledImage = imfill(binaryImage, 'holes'); % Fill all holes in all blobs.

Not sure why you're trying to compile a list of locations to fill. Please attach your restaurant.jpg image and I'll see what I can do.

[Walter Roberson]

Those are not individual locations to be filled: those are "seed" locations to start flood-filling. If the location happens to have a pixel set then nothing happens; if the location does not happen to be set then everything surrounding the connected component would be filled. This seems to be unlikely to be desired behaviour. I think the 'holes' option is much more likely to be the desired option.

[Image Analyst]

Yeah, that's what I meant.

And when he said he wants to "fill each connected component in the image" I thought "Why isn't he using the 'holes' option?"

[abdelrahim hashem]

I do not need to fill all connected components, just I need to fill some of them according to some conditions.

[Image Analyst]

What conditions? Explain in words since your code is uncommented and hard to follow. Your first for loop just seems to rebuild your original binary image, bw, which seems pointless.

Again, attach your image so we can try your code.

[abdelrahim hashem]

Now I knew the error; I have to convert the matrix of binary image into logical. When I convert it to logical, the next command has run:

BW = logical(BW);

BW2 = imfill(BW,XY_indices,8); XY_indices is the index of row and column of corner of all Bounding Boxes

[Image Analyst]

You should have done:

New_bw = false(size(g));

instead of

New_bw = zeros(size(g));

That said, the code is still weird, but anyway I know you don't want anymore help since you haven't attached your image after two requests, so I assume it's working to your satisfaction (quirky as it is), so good luck with the rest of the project.

Answer 3

[abdelrahim hashem]

[Image Analyst]

I'm not sure what you're trying to do. You get a Canny edge image but don't do anything with it. You binarize a gray scale version of this image, but what does that give you? Nothing special as far as I can tell. What objects in this image are you trying to find? Are all your images if this restaurant? Or do you have different scenes? Because some algorithm you make for this image might not work for another image. For example if you want the letters, you can find them by finding the largest blue blob and finding the holes inside that. However that might not find any letters if you're looking at a stop sign, or a white sign with black letters, etc.

[abdelrahim hashem]

Yes, this is part of a code, I should delete some statements from it like getting the Canny edge.

As regards my work I need to fill holes by white pixels in the letters only, this is sure after deletion non-characters objects, then I need to combine characters belong to the same word.

[Image Analyst]

Try the attached m-file.

[abdelrahim hashem]

Thanks, Image Analyst. I depended on the number of pixels in each region/objects in my previous work, Already I have published two papers in this direction, though I think that I need other properties/features to make good selection for text character regions and also to discard non-text character regions rather than number of pixels. Thanks again

[Image Analyst]

See https://www.mathworks.com/help/vision/examples/recognize-text-using-optical-character-recognition-ocr.html

and

https://www.mathworks.com/help/vision/ref/ocr.html#bt548t1-2_1