Segmentation of pictures into regions with a tile-by-tile method

摘要

A method is presented for segmenting a digital picture into regions of uniform value. Unlike most other methods, this method considers the image a piece at a time. Based upon the split-and-merge approach, the method is time-efficient. An important advantage over previous methods, however, is this method's space-efficiency. Large pictures are easily handled, since main memory need only accommodate a relatively small portion of the picture, together with the accumulated segmentation, which in practice can be easily represented much more compactly than the picture array can. Moreover, this ‘tile’-based approach organizes the computation of a segmentation in such a way that parallel processors can readily be applied to cut the time required. Also presented are ways to incorporate edge information into segmentation decisions when the primary local feature of interest is brightness. An implementation of the method and the results it produced are discussed.