Problem with cumsum

Hi guys,

I am trying to write a code to correct the background for an image. I used the formula given in the paper "Methods of Digital Video Microscopy for Colloidal Studies" by Crocker and Grier.

The formula is

Ib(x,y)=(constant)*[?I(x+i,y+j)] %the summation goes from i,j= -w to +w; w is predefined. 'I' is the original image and 'Ib' the background corrected image.

I thought of using "cumsum" since we need to sum up the components of the rows and columns of a matrix basically to decrease the computation time. However I am getting an error and I am clueless about what I am doing wrong.

The code is given below

I=imread('*.tif');

I=rgb2gray(I);

sizeim=size(I);                     %computes size of image

Iback=zeros(sizeim(1),sizeim(2));   %Iback same size as I but with zeros

B=1/((2*w+1)^2);                    %constant used for division

w=5;

for i=w+1:sizeim(1)-w

      for j=w+1:sizeim(2)-w

          temp=cumsum(cumsum(I(i-w:i+w,j-w:j+w),1),2);
          Iback(i,j)=temp(2*w+1,2*w+1)/B;

      end
  end

I get the error message "??? Undefined function or method 'cumsum' for input arguments of type 'uint8'"

Can anyone tell me what I am doing wrong? Your help is truly appreciated.

NS

Answer 1

[Matt Fig]

CUMSUM does not work for data that isn't single or double. You can write your own function to do this, or cast the data to single before passing to CUMSUM. Note however, that you will have to figure out what to do with overflow! The cumulative sum in uint8 may easily go over 255.

intmax('uint8')

You could also convert your image to have a datatype which has a larger range...

[Matt Fig]

NS says,

Thanks Matt. What do you mean by data being single or double?

[Matt Fig]

I mean the data type. You might want to read the help for CLASS. There are several data types in MATLAB, single, double, int32, uint32, int16, uint16, etc.

[NS]

Thanks Matt. I am a little more confused than I was before. Since my image is a 2D matrix, cumsum should work right?

Also, when I add in "I=double(I);" before the 'for loop', my code runs just fine. But I am not able to see my image at all. It is just a white background. How do I deal with this?

[Sean de Wolski]

The image appears to be all white because the range is assumed to be [0 1] for class double images. Thus everything 1 or greater appears white. Since it's not in the range [0 1], you have to tell matlab to show the whole thing.

imshow(I,[]);

[NS]

It worked. Thanks a lot Matt and Sean. :)

[Matt Fig]

To convert to double on [0 1], you need to keep the proportions.

I = double(I)/255;

And you image being 2D has nothing to do with the datatype! Don't just be confused, read the help for the CLASS function and read about datatypes in MATLAB.

[Sean de Wolski]

I don't usually bother with the conversion from 0-1. Are there any actual advantages to it?

[Matt Fig]

Perhaps not, I am not into image processing too much. I only think about the math of it that way. ;-)