Fig. 8.

2D illustration of cascaded feature calculations of macMIND. The “Multi sampling partners” sketch illustrates the msSSC with multi-sampling patterns (multi-scale sampling and multi-orientation sampling). msSSC includes some different scale self-similarity contexts (SSC). The left sketch illustrates a dual-scale SSC: the small-scale SSC includes the central patch P₀ (red box) and its closer 4-neighborhood (light blue boxes); the larger-scale SSC includes the central patch P₀ and its farther 4-neighborhood (dark blue boxes). Each SSC includes more connectivity (black lines and gray line) than MIND (gray lines), which leads macMIND to incorporate more orientation sampling. L and R₁ symbolize the patch distance and size, respectively. The msSSC feature map with M channels are created. The “feature aggregation” sketch shows the N bins (here N = 16) in log-polar space with 8-angle intervals and 2-radial intervals. One of the bins is colored with gray. macMIND translates each voxel in an image to a M × N matrix by macMIND. Finally, macMIND feature map is created as a M × N channel image for registration