Such processing maps from a world scene into an image.
Such processing maps from an image into one or more images.
Segmentation is a difficult and often ill-posed problem (in that it may not have an unambiguous solution). It is most often carried out in a domain-dependent way.
There are two primary ways to carry out segmentation: by finding boundaries between distinct regions, or by finding the regions that are separated by boundaries. These two ways of looking at the same problem are closely related, but are achieved through very different mechanisms.
Segmentation maps from images into images plus segments.
For example, character representation tasks would employ representations and descriptions of features such as lakes and bays in written glyphs. Military reconnaisance systems might associate textures with regions.
This kind of processing takes images and segments and maps them into images, segments, and descriptors.
In a broader sense, the sollipsist and proponent of strong AI might say that there is no meaning beyond form, that all semantics is nothing but another form of syntax. Most people, however, feel there is a distinction between form and meaning, thus we will act as if such a distinction exists.