TABLE OF CONTENTS We now present various aspects of source modeling.
In what follows, we are exclusively concerned with discrete sources, i.e. sources which send information in the form of discrete symbols (samples, impulses, 0/1), as this corresponds to the general case in the telecommunications field. We distinguish the case of sources with memory and without memory. Emphasis will be placed in this context on the general model of the discrete Markov source (with memory). We show how we can approximate such sources with memoryless sources, by introducing the notions of extension and adjoint source. Then, as an example, we present the two main types of sources: images and texts.
A source is called memoryless when the occurrence probability of a symbol does not depend on the symbols previously observed. Denoting as [ S] = [ s i, s 2, , s k] the set of source symbols, we then have p( s i n s i n?1, s i n?2, ) = p( s i n), index n standing for the rank of the symbol in the sequence.
On the other hand, a source is said to be with memory if the relation does not hold.
Studying this kind of source is mainly a matter of Markov graphs; these sources are called Markov sources. The source evolution is fully determined by its initial state...
