On Jun 9, 5:43=A0pm, AJ <arandal...@[EMAIL PROTECTED]
> wrote:
> On Jun 9, 10:30 pm, aruzinsky <aruzin...@[EMAIL PROTECTED]
> wrote:
>
>
>
>
>
> > Interspersed comments:
>
> > On Jun 9, 11:57 am, AJ <arandal...@[EMAIL PROTECTED]
> wrote:
>
> > > I'm impressed! =A0The edges are very crisp, much better than I am
> > > currently getting. =A0Thank you for going to the effort of
experimenti=
ng
> > > with it and uploading your results.
>
> > > It would be great if I could use my results (iterative restoration
and=
> > > interpolation) and then apply your technique to edges, as I think
that=
> > > the "inner" details such as the face look quite good with my
> > > restoration technique, and the edges of the image look much better
> > > with yours.
>
> > The edge preserving method is not post processing and is proprietary.
>
> > > I would be really grateful if you could expand on the process you
took=
> > > to achieve this image, such as the "8X nearest neighbor (box)
> > > enlargement".
>
> >http://en.wikipedia.org/wiki/Nearest_neighbor_interpolation
>
> > I didn't produce any box enlargement. =A0I pretended that your image
was=
> > a box enlargement of a hypothetical image 1/8 its size. =A0 Thusly,
only=
> > for my convenience, I threw away all information contained in your DCT
> > coefficients above 0 frequency. =A0With respect to reduction, an
> > interpolation kernel is typically expanded by a factor inversely
> > pro****tional to the scale. =A0More specifically, the pixels in each of
> > your 8x8 blocks were simply averaged to produce one pixel in a reduced
> > image. =A0Then that reduced image was enlarged by my proprietary
method
> > which incor****ates prior knowledge that the image was a box reduction.
>
> > >Did you implement this in Matlab? =A0I'm not sure what
> > > you mean by this...is it a filter you have applied?
>
> > No, I implemented SAR Image Processor in C++. =A0You can download a
free=
> > save disabled demo to play with. =A0Google it.
>
> Thanks for your reply.
>
> I did find SAR Image Processor after I had responded to your last
> message. =A0However, I am running a Linux OS so unfortunately cannot run
> it.
>
> I did however use interpolation to improve the quality of my image, so
> thank you for that suggestion. =A0I have uploaded my "best" attempt at
> restoring the image, and it can be seen at:
>
> http://i287.photobucket.com/albums/ll146/arandall85/interpolated_iter...
>
> To get to this point, I have used blind deconvolution, and then
> interpolated the result. =A0I'm quite happy with this outcome, although
> it is a little blurry. =A0What do you all think?
>
> Ideally, I would like to "fix" the edges in some way, as they are
> still a bit blocky, and I would like to sharpen the whole image
> slightly - but I think I may be hoping for a bit too much! ;)
>
> If anyone has any further suggestions to improve this image, I would
> be really grateful. =A0Nearly there...
>
> Kind regards,
>
> AJ- Hide quoted text -
>
> - Show quoted text -
But, although similar, this is not a deconvolution problem. Like
convolution, you have a linear transformation which can be put in
matrix form,
Ax =3D y
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. In low rank cases, it is generally good
practice to constrain a solution, x', so that Ax' =3D y. I checked your
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. If
you check my result, you will see that the averages are equal to those
of the input image. You also want the non-zero frequency DCT
coefficients for the results to equal those of your input image,
which, of course, my result violated.
As an experiment, I added the non-zero frequency DCT components to my
previous result and got,
http://www.general-cathexis.com/images/degradedimagesh2_3.png
, which is one of an infinite number of possible solutions, x', for
Ax' =3D y.
|
