|
| |
| The basis of any image processing problem is a
semantics: a collection of propositions one might want to assert about the
images with which one is dealing, together with criteria for the
verification of the correctness of these propositions if asserted about a
particular image. A simple example is the semantics consisting of the
propositions 'the volume in the world that projected to region R in the
image contained a human being in the foreground', together with the
ability of human beings to verify the correctness of this statement upon
looking at any particular image. The job of image processing or computer
vision algorithms is to take images as input and to assert propositions from
the semantics that are correct according to the criteria as output. A
semantics is both necessary and sufficient for a well-defined problem. The
absence of a semantics creates great difficulties since evaluation is
impossible: examples include unsupervised segmentation and query by
example.
|
|
