Notes on Fiji Analyze > Colocalize Threshold
1.52p Java 1.8.0_172
20190906

Both images must be 8 or 16 bit grayscale. If the second channel is 32 bits, an error will not be thrown but the results will be wrong.
Zero should be used for masking, not NaN.

This does not work with ROIs. Therefore, need to duplicate portions of image for analysis and clear outside with zeros turned off inthe calculation. Because the thresholding is automated

In the following example, a square 10 X 10 pixels in two channels started completely overlapped. The background value = 1; if it were zero, the results would be different. The square was iteratively shifted one pixel in the X axis for each measurement. This meant that there was a 10% reduction of overlap. In the table below, look at the first column and the third column (Rtotal) for Pearson's coefficient.

These Rtotal results correspond to other software for calculating Pearson's correlation.

Links to macro and test dataset.