Image processing, DCT coefficients ordering

Hi guys,

I am currently doing a project about the multiple description coding for a
image using correlating transform. In one of the IEEE journals, the author
describes the encoder as:

“The coder is based on a modification of a standard JPEG coder. In our
baseline encoder, each block of samples is transformed using an 8*8
discrete cosine transform (DCT). These coefficients are ordered to have
decreasing variance.”

I have done the DCT part, but can’t figure out how to rearrange these
coefficients in the orders of decreasing variance. I thought the variance
is only meaningful when it refers to a set of random variables and the
value of variance will not be affected by the ordering of variables.

Could anyone help me with this problem? Thank you very much.

On Jun 26, 12:22 pm, "destiny_nz" <fanzuw...@hotmail.com> wrote:
> Hi guys,
>
> I am currently doing a project about the multiple description coding for a
> image using correlating transform. In one of the IEEE journals, the author
> describes the encoder as:
>
> "The coder is based on a modification of a standard JPEG coder. In our
> baseline encoder, each block of samples is transformed using an 8*8
> discrete cosine transform (DCT). These coefficients are ordered to have
> decreasing variance."
>
> I have done the DCT part, but can't figure out how to rearrange these
> coefficients in the orders of decreasing variance. I thought the variance
> is only meaningful when it refers to a set of random variables and the
> value of variance will not be affected by the ordering of variables.

Indeed.  The variance is calculated across all images, and the
reordering order is fixed, and part of the JPEG spec.  See e.g.
http://en.wikipedia.org/JPEG.


--
Oli

Hi,
  I am also working on Multiple Description Coding. 
>>
>I have done the DCT part, but can’t figure out how to rearrange these
>coefficients in the orders of decreasing variance. I thought the
variance
>is only meaningful when it refers to a set of random variables and the
>value of variance will not be affected by the ordering of variables.
>
>Could anyone help me with this problem? Thank you very much.
>
  Probably here each of the coefficeints is taken as a random variable
with similarly positioned coeffs in all the 8 x 8 blocks being its values.
For example if you have a 512 x 512 image, you will have 4096 8x8 blocks .
Say the coeff in (1,1) is a ramdom variable then it will have 4096 sample
values. hope I am clear. 
  BTW, How do simulate sending MDs across multiple channels in Matlab?
REgards,
Chakrala