Forums

Vector Quantization: Problem with designing codebook

Started by Jake December 30, 2005
> > GO TO THE LIBRARY > > VLV >
Great answer.
Jake wrote:
>>GO TO THE LIBRARY >> >>VLV >> > Great answer.
No, seriously -- you need to read up a little bit a good description of Vector Quantization. Remember that you replace *an entire vector* by an index in the codebook. Notice also that Vladimir did mention that you split the vectors into several components (I also suggested that earlier in the thread -- didn't your newsreader show my previous reply?). That is, if you have a 30-dimensional vector, then you can split it into three vectors, 10-dimensional each; now, you can use a codebook of, say, 8 bits, to encode each of the three vectors. The result is that you're using a total of 24-bits to encode each 30-dimensional vector. If the above does not clarify, then *really* you need to hit the books first, to get a general idea of the technique, and then we can give you a hand with any specific doubts you may have. HTH, Carlos --
Jake wrote:
>>GO TO THE LIBRARY >> >>VLV >> > > > Great answer. > >
If you don't have enough time to go to library, you may want to refer the following paper: Vector quantization Gray, R.; ASSP Magazine, IEEE [see also IEEE Signal Processing Magazine] Volume 1, Issue 2, Part 1, Apr 1984 Page(s):4 - 29 Sorry if it was already mentioned. -James Gold