You may want to consider the similarity distance between a pair images or between a pair of regions-of-interest in a pair of images. Here is an approach to doing this.

Let X, Y be sets of points in a pair of images. In this context, a picture point can be thought of as a place with a location and having measurable content such as colour brightness or greylevel intensity or gradient orientation. Let $\Phi$ be a set of probe functions used to extract feature values from each picture point. Let th > 0 be a threshold. Then compute

\begin{align*}

D(X,Y) = inf\{d(\Phi(x),\Phi(y)): x\in X, y\in Y\},\mbox{where}\\

d(\Phi(x),\Phi(y)) = \mbox{cityblock distance between feature vectors}\ \Phi(x),\Phi(y).

\end{align*}

Here, the formula for similarity distance is written in LaTeX.