We will end up having a matrix of integers the with the same columns of the original matrix but only k rows. Each vector contains a ranked list of the k stronger pitch classes, sorted in descending order. Note the underlying assumption is that in order to represent and summarise a chroma vector, the stronger a class energy is, the better.
Almost done: the Interval Matrix is obtained applying the Intervals Table as a function between cells (ranks) having the same index and belonging to subsequent chroma vectors.
In this way we compute a distance between adjacent vectors (x[n] – x[n1]) and those measures compose an output matrix that has:

 the same number of rows of the input matrix
 m1 columns: where m is the number of the columns of the input matrix