On Jun 10, 2:46=A0am, Martin Brown <|||newspam...@[EMAIL PROTECTED]
>
wrote:
>
> > and you want to estimate x from y, but, unlike convolution, most of
> > the eigenvalues of A are zero and (I am uncertain so check me) A is
> > idempotent, i.e., A^2 =3D A. =A0 In low rank cases, it is generally
good=
> > practice to constrain a solution, x', so that Ax' =3D y. =A0I checked
yo=
ur
> > result and it violates this constraint because the averages of the 8x8
> > blocks in your result are not equal to those of your input image.
>
> You shouldn't expect them to be exactly equal.
I didn't. For my result, the RMS of differences in block averages was
0.042 before 8 bit quantization and 0.29 after 8 bit quantization.
For the OP's result, the RMSE is 6.35. For engineers, that's equal
vs. unequal.
> That is overconstrained.
That is under-constrained.
> The best you can hope for is that the DC and First AC component (which
> is all he has) computed from the model image are on average within half
> the quantisation error of the JPEG encoding in a least squares test.
>
Who said it was JPEG encoding which uses gross quantization for lossy
compression? In the absence of an explicit statement of the OP that
it is JPEG encoding, it is reasonable to assume that the only
quantization error is 8 bit and, unless you examined the OP's PNG
format, for all you know, it could be 16 bit. Eight bit quantization
noise is invisible and computationally negligible in this case.
|
