how does correlation-along-space be done in the FMRI dissimilarity matrixs?

I'm researching on dissimilarity matrices used for interpretation of FMRI signals taken from brain. In the following image, it has been stated that the correlation (1 - correlation) has been calculated for when each of the image have been shown to the person. It has written correlation along space. How is correlation along a space calculated. Is the image for one instance of time or along time? Is it for one voxel or an area of voxels? How are the areas chosen? If anybody can explain thoroughly I would really appreciate it. If after answering to my question anybody could also explain (or give a link) how MVPA (multi-voxel pattern analysis) is done on the data, all my initial questions will be answered. This is the link to the image that tells everything