To calculate images similarity based on Mutual Information (MI), histogram and join histogram is required. In the literature, there are 2 contradict opinions about MI operation. 1. MI does not utilize image spatial information to calculate the similarity. 2. MI is embedded spatial information indirectly by joint histogram. I thought that may I misunderstand the MI methodology.
... mutual information (MI) considers only statistical information, while spatial information is completely neglected. (Abdel-Basset et al. 2017)
Several works (Gao et al., ..) have adopted the joint histogram of the image pair to embed spatial information into an intensity distribution. (Xu et al. 2016)
I believe the second opinion is the right one. But I find many papers agree with the first one. Which of the two groups do you believe in? why is another group wrong?