Elias gamma coding

"Gamma encoding" redirects here. For the signal processing operation, see gamma correction.

Elias gamma code is a universal code encoding positive integers developed by Peter Elias.[1]:197, 199 It is used most commonly when coding integers whose upper-bound cannot be determined beforehand.

Encoding

To code a number x≥1:

  1. Let N=⌊log2 x⌋ be the highest power of 2 it contains, so 2N ≤ x < 2N+1.
  2. Write out N zero bits, then
  3. Append the binary form of x, an N+1-bit binary number.

An equivalent way to express the same process:

  1. Encode N in unary; that is, as N zeroes followed by a one.
  2. Append the remaining N binary digits of x to this representation of N.

To represent a number x, Elias gamma uses 2 \lfloor \log_2(x) \rfloor + 1 bits.[1]:199

The code begins (the implied probability distribution for the code is added for clarity):

Number Binary γ Encoding Implied probability
1 = 20 + 0 1 1 1/2
2 = 21 + 0 1 0 0 1 0 1/8
3 = 21 + 1 1 1 0 1 1 1/8
4 = 22 + 0 1 00 00 1 00 1/32
5 = 22 + 1 1 01 00 1 01 1/32
6 = 22 + 2 1 10 00 1 10 1/32
7 = 22 + 3 1 11 00 1 11 1/32
8 = 23 + 0 1 000 000 1 000 1/128
9 = 23 + 1 1 001 000 1 001 1/128
10 = 23 + 2 1 010 000 1 010 1/128
11 = 23 + 3 1 011 000 1 011 1/128
12 = 23 + 4 1 100 000 1 100 1/128
13 = 23 + 5 1 101 000 1 101 1/128
14 = 23 + 6 1 110 000 1 110 1/128
15 = 23 + 7 1 111 000 1 111 1/128
16 = 24 + 0 1 0000 0000 1 0000 1/512
17 = 24 + 1 1 0001 0000 1 0001 1/512

Decoding

To decode an Elias gamma-coded integer:

  1. Read and count 0s from the stream until you reach the first 1. Call this count of zeroes N.
  2. Considering the one that was reached to be the first digit of the integer, with a value of 2N, read the remaining N digits of the integer.

Uses

Gamma coding is used in applications where the largest encoded value is not known ahead of time, or to compress data in which small values are much more frequent than large values.

Gamma coding is a building block in the Elias delta code.

Generalizations

Gamma coding does not code zero or negative integers. One way of handling zero is to add 1 before coding and then subtract 1 after decoding. Another way is to prefix each nonzero code with a 1 and then code zero as a single 0.

One way to code all integers is to set up a bijection, mapping integers (0, −1, 1, −2, 2, −3, 3, ...) to (1, 2, 3, 4, 5, 6, 7, ...) before coding. In software, this is most easily done by mapping non-negative inputs to odd outputs, and negative inputs to even outputs, so the least-significant bit becomes an inverted sign bit:
\begin{cases} x \mapsto 2x+1 & \mathrm{when~} x \geq 0 \\ x \mapsto -2x & \mathrm{when~} x < 0 \\ \end{cases}

Exponential-Golomb coding generalizes the gamma code to integers with a "flatter" power-law distribution, just as Golomb coding generalizes the unary code. It involves dividing the number by a positive divisor, commonly a power of 2, writing the gamma code for one more than the quotient, and writing out the remainder in an ordinary binary code.

References

  1. ↑ 1.0 1.1 Elias, Peter (March 1975). "Universal codeword sets and representations of the integers". IEEE Transactions on Information Theory 21 (2): 194–203. doi:10.1109/tit.1975.1055349.

See also