The images containing only blocks, and things made from blocks, are but a small and unnatural subset of the set of all images. It is essential to be able to treat of more general objects before we can feel warranted in taking ourselves seriously. So the question arises, can we decompose more complicated objects than linear images? Happily the answer is that we can. If we have an image built up out of curves, at some suitable level of resolution, between where the noise intrudes and the global structure intrudes, we may use the same process to perform chunking, and the same process to perform the UpWrite of each chunk. Higher order moments than two will generally be necessary in order for the general shape of the curved chunk to be described with sufficient precision to allow a convincing DownWrite. There is plainly, however, no intrinsic obstruction to dealing with images composed of piecewise curved pixel segments. Moreover, the resolution radius is a parameter which can be adjusted in some cases to distinguish noise from signal, a point which will be enlarged on later.
The significance of curved elements is, as the thoughtful reader will have noted, that they arise rather naturally from physical objects coming in front of other physical objects. Boundaries can often be taken to be piecewise curved objects with noise.
There are various ways of chunking into curves: the program recog which can be found on the site:
ftp://ciips.ee.uwa.edu.au/pub/syntactic/c_code/rht/shows a variety of methods.