The problem of finding appropriate mathematical objects to model image
s is considered. Using the notion of a completed graph of a bounded fu
nction, which is a closed and bounded point set in the three-dimension
al Euclidean space R(3), and exploring the Hausdorff distance between
these point sets, a metric space IM(D) of functions is defined. The ma
in purpose is to show that the functions f is an element of IM(D), def
ined on the square D = [0, 1](2), appropriate mathematical models of r
eal world images. The properties of the metric space IM(D) are studied
and methods of approximation for the purpose of image compression are
presented. The metric space IM(D) contains the so-called pixel functi
ons which are produced through digitizing images. It is proved that ev
ery function f is an element of IM(D) may be digitized and represented
by a pixel function p(n), with n pixels, in such a way that the dista
nce between f and p(n) is no greater than 2n(-1/2). It is advocated th
at the Hausdorff distance is the most natural one to measure the diffe
rence between two pixel representations of a given image. This gives a
natural mathematical measure of the quality of the compression produc
ed through different methods.