inpainting_nans by John D'Errico ERROR MESSAGE

I used the inpainting_nans that is available here http://mathworks.com/matlabcentral/fileexchange/4551

I used an image file with a black background picture and a white spot on the middle of it. A = imread('F:\image.bmp');

and i used the inpainting_nans code to see what will be its effect B = inpaint_nans3(A);

But I encountered this error:

?? Error using ==> mtimes Sparse integer array arithmetic operations are not supported.

Error in ==> inpaint_nans3 at 159 rhs=-fda(:,known_list)*A(known_list);

What does this mean?

Answer 1

[Titus Edelhofer]

Hi, you will probably have to convert the image from integer values to doubles. Try

Ad = double(A);

and run inpaint_nans with Ad ...

Titus

[Sean de Wolski]

Yes. inpaint_nans requires conversion to double. You could also just add these lines to the beginning:

if ~isa(A,'double')

A = double(A);

end

[Paulo]

thank you very much to the both of you

[DGM]

I should add that if your image is integer-class (e.g. uint8), then it has no NaNs. Casting to 'double' will eliminate the error, but unless you do something to add NaN pixels to the image, there will be nothing for inpaint_nans to fill.

A = imread('cameraman.tif'); % uint8
A = im2double(A); % double, correctly-scaled for its class

% there's nothing to do here
op1 = inpaint_nans(A);

imshow(op1) % the image is unchanged

The above example accomplishes nothing, since there's nothing to inpaint. You'll have to create a mask from something and use that to apply NaNs to the image regions that need to be estimated.

% a logical mask
mk = imread('cmantifmk.png')>128;

% use NaNs to embed the mask in the image
A(mk) = NaN;

% there's actually something to do now
op2 = inpaint_nans(A);

imshow(op2)

... not that that's a good mask to demonstrate inpainting with, but it at least shows that inpainting is happening.