regionprops 0 Area NaN Centroid issue

I am using "regionprops" for image analysis. It generally works well, but for 1 image now "regionprops" is returning an area of zero.

Before "regionprops" I threshold the image and then use "bwlabel".

rp = 

struct with fields: 

Area: 0 
Centroid: [NaN NaN] 
BoundingBox: [0.5000 0.5000 0 0]

Do you understand this? How can a region have area 0 (=no pixel)?

Not sure if this is related, but for memory issues my labelled image is uint16 and not double.

Answer 1

[MathWorks Support Team]

This expected behavior for "regionprops". Please see the explanation below for more information.

You can reproduce this behavior by using a labelmatrix. Any image that is not binary or a bwconncomp structure is considered as Labelmatrix by "regionprops". When you use an uint8 (for example) image as input, each integer value is considered as marking a different region. If some integer values are missing, there’s still a field created for that region but "regionprops" returns this structure with Area 0 & Centroid NaN.

% The test image below image has a max of 5 labels.
% Regionprops expects to find 5 regions labelled 1 to 5.
% Since label ‘2’ is missing in the image, regionprops will treat that as an empty region.

>>A =   [ 1     0     0     0     0     3     0     4     0     5;
          1     0     0     0     0     3     0     4     0     5;
          1     0     0     0     0     3     0     4     0     5;
          1     0     0     0     0     3     0     4     0     5;
          1     0     0     0     0     3     0     4     0     5];

>>F = regionprops(A)

F = 

    5×1 struct array with fields:

      Area
      Centroid
      BoundingBox

% Querying the properties of the empty region labeled as ‘2’
>> F(2)

ans = 

    struct with fields:

             Area: 0
         Centroid: [NaN NaN]
      BoundingBox: [0.5000 0.5000 0 0]

[Image Analyst]

You can relabel the binary image if you removed something from the labeled image, like what they did

A = [...
	1     0     2     0     0     3     0     4     0     5;
	1     0     2     0     0     3     0     4     0     5;
	1     0     2     0     0     3     0     4     0     5;
	1     0     2     0     0     3     0     4     0     5;
	1     0     2     0     0     3     0     4     0     5];
A(A==2) = 0 % Get rid of 2 from labeled image and make it zero.
% Now there is no 2 in the labeled image and you'll get nans for measurements from the missing #2.
props = regionprops(A)
% Show what we get for the missing 2:
props(2)

% Now the fix.  Simply relabel the labeled image
binaryImage = A~= 0           % Create a new binary image.
newA = bwlabel(binaryImage)   % Give new labels to that new binary image.
% newA =
%      1     0     0     0     0     2     0     3     0     4
%      1     0     0     0     0     2     0     3     0     4
%      1     0     0     0     0     2     0     3     0     4
%      1     0     0     0     0     2     0     3     0     4
%      1     0     0     0     0     2     0     3     0     4

% Get new props.  props will have only 4 structures this time instead of 5.
props = regionprops(newA) % Use newA this time, not the original A
% Show what we get for the 4 blobs:
props(1)
props(2)
props(3)
props(4)