Vector Quantization

Image quantization is termed as the process of mapping the continuous set of pixels. The encoder mapping performed in the image quantization is termed as the image quantization process. The decoder mapping performed in the image quantization is termed as the inverse image quantization.

There are two kinds of image quantization namely “scalar quantization” and “vector quantization”.

In scalar quantization, each sample or pixel is quantized independently. A scalar quantizer maps a continuous probability density function into a discrete set of reconstruction levels through the set of decision levels mapping the continuous set of pixels.

In vector quantization, each pixel is not quantized, but instead, a set of continuous pixels are represented as vector through the vector states.

How it works?

As the vector quantization is the set of continuous pixels represented as set of vectors through vector states, where the set of vectors are termed as "codebook" in both encoder as well in decoder level. Each vector in the codebook will be assigned a binary index.

(a) Encoder level:
Step 1--- Compressing into code blocks.
Step 2--- Generation of codebook for the code blocks.
Step 3--- Eliminating data redundancy.
Step 4--- Optimized codebook generated.
Step 5--- Transformed into code vectors from indices.

(b) Decoder level
Step 1--- Reverse process is applied.
Step 2--- Reconstruction of data into original format.


The vector quantization can be classified into:-
(i) Memoryless Vector Quantization
And
(ii) Vector Quantization with Memory

Starting with Memoryless Vector Quantization:- The quantization is being performed independently with the help of the most preferred algorithm "K-means clustering algorithm". But there is something better than this Memoryless vector quantization. The hidden markov comes into the game where it does provide in giving finite set of possibilities, in which the possibilities can be extended from one state to any state. This also means that it doesn't totally rely on the previous one followed in memoryless vector quantization (one-to-one transition). In web performance, it means:- The data must be steadily transmitted inspite of one byte loss, then the next byte should go ahead. (i.e.) Bytes must not be dependent over eachother during the independent transfer.