Images similarity measure using j-divergence

I need to reproduce the result of Xu, Xiaocong, et al, 2016 paper . Mutual information code already was done and work fine. can anyone help me to write Jeffrey’s divergence code? because the I use file exchange code but I think it has an issue.

I the aforementioned paper the j-div is better because it has wider distance between correct result and wrong one, but in my case, the opposite MI is better (see the image below)

(up paper result, down my result)

Thanks