Академический Документы
Профессиональный Документы
Культура Документы
Encoder Decoder
Input Output
Vector Vector
Find Closest Code
Table Lookup
Vector
• Quantization function
6. VQ is used for low-bit rate applications where low 6. In image compression, SQ creates annoying effects
resolution is sufficient. It is widely used for image in decompressed image.
compression.
7. For a given rate, VQ results in lower distortion than 7. For a given rate, SQ results in greater distortion
SQ. than VQ.
LBG (Linde-Buzo-Gray) Algorithm
• It is used to design codebook.
• Most widely used approach.
• Set of quantizer output points are called codebook of quantizer.
• The process of placing these output points is referred as codebook
design.
• It is based on K-means algorithm (Clustering procedure).
• It guarantee that the distortion from one iteration to the next will not
increase.
Cont…
• K-means Algorithm:
Large set of output vectors (training set).
Initial set of K representative vectors (representative set)
The representative set is updated by computing the centroid of the training
set vectors which are assigned to it. (by mean value of all the vectors assigned
to representative vectors)
Cont…
• LBG Algorithm:
Choose initial set of code vectors ri and tolerance value e.
m=1; D0 = ∞;
While not success
Divide the set M training vectors x into L clusters Ki using the minimum
distortion condition:
x ∈ Ki if d(x, ri) ≤ d(x, rj) for all j≠i, 1 ≤ I ≤ L;
Compute the average distortion
r0 r1
where mi’s are arbitrary integers and ui’s form a linearly independent
set of n<=k vectors
• Example of lattice quantizer in two dimensions: Hexagonal lattice.
• Low complexity
• Codebook storage is eliminated
• Gives lower distortion for the same rate, compared to a uniform scalar
quantizer.
Hexagonal Lattice
Thank You