why am i getting NaN from my function

i was asked to write a function that gets b&w image and returns its entropy. im not allowed to use im_hist or entrtopy (obviously).

this is the function i wrote:

function [Entropy] = EntropyCalc(image)

image_hist = histogram(image);

values = image_hist.Values;

[size_x, size_y] = size(image);

total = size_x * size_y;

p = values./total;

Entropy = - sum(p.*log2(p));

end

but i keep getting NaN.

any ideas??

thanks!

[David Goodmanson]

Hi tomer, don't forget 0*log(0) = nan

[tomer orenshtein]

thanks! missed that

Answer 1

[Thiago Henrique Gomes Lobato]

As David said your log2(p) is not safe, you can solve this problem by using this line of code:

Entropy = - sum(p.*log2(p+eps));

[tomer orenshtein]

thank you! that solved that. now i need to find out why the answer is different than the built in entropy function.

thanks!

[Thiago Henrique Gomes Lobato]

The difference is in the histogram function, you need to explicitely define the number of bins to 256 (make sure image is uint8) so it has the same result.

 image_hist = histogram(image,256);

Or faster:

values  = imhist(image(:));

[tomer orenshtein]

that was the thing... didnt know this option. (cant ust imhist in this hw). thanks again! helped me a lot!