how to compare ground truth image with segmented image?

BW2=imread('C2.png'); %segmented image

figure,imshow(BW2)

BW=imread('001_mask.png'); %groundtruth-image

figure,imshow(BW)

% mask = false(size(BW2));

% mask(25:end-25,25:end-25) = true;

% BW = activecontour(BW2, mask, 300);

similarity = jaccard (BW2, BW)

figure

imshowpair(BW2, BW)

title(['Jaccard Index = ' num2str(similarity)])

I am trying to find accuracy but got an error.

Undefined function 'jaccard' for input arguments of type 'uint8'.

Error in Untitled (line 8)

similarity = jaccard (BW2, BW)

what will be changed in this code?

[Adam]

Do you have the Image Processing Toolbox and are you using R2017b or later (it does help a lot if you fill in the Products and Release section of the question)? What does

which jaccard

show?

[Arslan Ahmed Awan]

You have to Install MATLAB R2017b because jaccard was introduced in Image Processing Toolbox of R2017b.

Answer 1

[Constantino Carlos Reyes-Aldasoro]

Jaccard index is the intersection (BW.*BW2) over the union (BW|BW2) or ((BW+BW2)>0), so you can calculate directly like this

sum(sum( BW.*BW2)) / sum(sum( BW|BW2))

Accuracy is different, that is how many pixels are exactly the same (including background)

sum(sum(BW==BW2)) / prod(size(BW))

Answer 2

[Leandro Hidalgo Torres]

Hi,

Jaccard needs their arguments to be logical type. Check the data type of BW and BW2, surely they are uint8 type. If it's true, use BW=logical(BW) and the problem will be solved,

Best regards,

LH