Figure 3.

Context features. (A) Relative context cue locations cp in the global coordinate system xo, yo, zo are rotated according to the orientation estimate of the voxel of interest ni to yield locations cℓip that are consistent. (B) At each of these locations, image channels are summed over cubes of radius r around their center. Our approach employs AdaBoost to select the most discriminative features for synapse segmentation.