sparse image representation via combined transforms - Convex ...
sparse image representation via combined transforms - Convex ...
sparse image representation via combined transforms - Convex ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
2.1. IMAGE CODING 7<br />
2.1.2 Source and Channel Coding<br />
After reviewing Shannon’s abstraction of communication systems, we focus our attention on<br />
the transmitter. Usually, the decoding scheme at the receiver is determined by the encoding<br />
scheme at the transmitter. There are two components in the transmitter (as shown in Figure<br />
2.2): source coder and channel coder. For a discrete-time source, without loss of generality,<br />
the role of a source coder is to code a message or a sequence of messages into as small a<br />
number of bits as possible. The role of a channel coder is to transfer a bit stream into a<br />
signal that is compatible with the channel and, at the same time, perform error correcting<br />
to achieve an arbitrarily small probability of bit error.<br />
TRANSMITTER<br />
SOURCE<br />
CODER<br />
CHANNEL<br />
CODER<br />
MESSAGE<br />
BITS<br />
SIGNAL<br />
Figure 2.2: Separation of encoding into source and channel coding.<br />
Shannon’s Fundamental Theorem of communication has two parts: the direct part and<br />
the converse part. The direct part states that if the minimum achievable source coding<br />
rate of a given source is strictly below the capacity of a channel, then the source can be<br />
reliably transmitted through the channel by appropriate encoding and decoding operations;<br />
the converse part states that if the source coding rate is strictly greater than the capacity of<br />
channel, then a reliable transmission is impossible. Shannon’s theorem implies that reliable<br />
transmission can be separated into two operations: source coding and channel coding. The<br />
design of source encoding and decoding can be independent of the characteristic of the<br />
channel; similarly, the design of channel encoding and decoding can be independent of the<br />
characteristics of the source. Owing to the converse theorem, the reliable transmission is<br />
doable either through a combination of separated source and channel coding or not possible<br />
at all—whether it is a joint source channel coding or not. Note that in Shannon’s proof, the<br />
Fundamental Theorem requires the condition that both source and channel are stationary<br />
and memoryless. A detailed discussion about whether the Fundamental Theorem holds in<br />
more general scenarios is given in [138].<br />
One fact I must point out is that in statistical language, Shannon’s theorem is an