Fig. 3. Vector voting, distributed block alignment, and composition with decay.

a Sequential alignment with vector voting. During block alignment sections are aligned sequentially to the already aligned portion of the stack. Vector voting is performed by aligning the source image to multiple target images, and passing the resulting fields through a pixel-wise smooth median computation. b Producing block stitching fields. To produce global alignment, the dataset is broken into overlapping blocks, which are aligned independently. The overlapping regions are aligned to each other to produce a stitching field between each two neighboring blocks. c Block stitching. The blocks are stitched together by applying each stitching to successive sections, while reducing its influence with a distance based weight decay.