When you convolve two matrices (images) of different sizes you get an output image that has row dimension equal to the sum of the two input matrices' row dimensions - 1. The same is true for the column dimension. The column dimension (size) of the output matrix is the sum of the column dimensions of the two input matrices - -1.
In your case, b and Ksigma are your two input matrices.
For example:
X = randn(3,3);
Y = randn(3,2);
Z = conv2(X,Y);
The output has row size 3+3-1 and column size 3+2-1
Using the 'same' option says only return the "central" part of Z in the above that matches the size of the first input image.
Z1 = conv2(X,Y,'same');
The output has the same size as X. If you compare Z1 and Z, you'll see what is in common.
