Science and Technology news... :: View topic

server · Guest

message unavailable

Aatu · Guest

*Yleisön pyynnöstä alan julkaisemaan Suomen ydinaavikoitumisen
nykytilastamme kertovaa faktaa.
M.T.05.05-08. Ilmatieteen laitos.

PAIKKA SADE mm.
_________________
Helsinki
Kiikkala
Turku 0,2mm
Jormala
Pori
Niinisalo
Tampere 1,5
Jokioinen 3,7
Lahti
Utti

L.ranta
Mikkeli
Ilomantsi
Joensuu
Kuopio
Viitasaari
Jyväskylä
Ähtäri
Kauhava
Vaasa 0,4

Ylivieska
Kajaani
Ruukki
Pudasjärvi
Suomussalmi
Kuusamo 0,1
Rovaniemi
Pello 1,7
Salla 0,1
Sodankylä 1,4

Muonio 0,1
Kilpisjärvi 1,0
Ivalo 0,3
Utsjoki 0,1

*KESKIARVO= 34kpl/ 0,31mm
Keskisadantavrk. 2mm/5vrk/10mm= 3,1%

JÄI -96,9% satamatta
_________________

*Aika huimaavat kertymät siis jälleen kerran ydinaavikoituvassa
maailmassamme. Toki voin tuoda esiin, että tällaisen tiedon jälkeen
totaalivedätystään korostaakseen samainen Ilmatieteen laitos tietoisena
siitä, ettei maassamme ole satanut liki 2kk aikaan vedättää pokkana, että:
"Viime aikoina on satanut tavannomaista enemmän. Kuukausisadannan
vaihteluväli oli ollut 20- 60mm välillä". Totuus tuosta kun on se, että
maassamme pitäisi normisataa 2mm/vrk. Eli edelläkerrottu on 33%-100%
normaalistamme. Ilmatieteen laitos valehtelee julkisesti 67% tiputtelun ja
ydinaavikoitumisellemme nykytyypillisen -33% vesitakadon olevan mielestään
NORMAALIA! Lisäksi kaiken huipuksi väittää nykytilanteen olevan
"tavannomaista vetisemmän ja enemmän!" Siltapaa ja ydinalan käskytyksen
mukaan näyttää tämä valtiolaitos etenevän kansallishuijauksissaan.

bite · Guest

On 27 Giu, 19:51, PeterOut <MajorSetb...@excite.com> wrote:
[quote]On Jun 27, 1:25 pm, Martin Leese <ple...@see.Web.for.e-mail.INVALID
wrote:

PeterOut wrote:
If you have a set of points belonging to 2 or more intersecting lines
in 2D, what is the best way to fit the lines to the points. Would it
be the Hough transform?

Is this a class assignment? If so, say so.
If not, describe your problem (where do the
points come from).

--
Regards,
Martin Leese
E-mail: ple...@see.Web.for.e-mail.INVALID
Web:http://members.tripod.com/martin_leese/

No. I finished with university classes 14 years ago. For my
particular problem, there are a number of reasons why I may not be
more specific about the source. I can say that they are fairly clean
(as opposed to random) points on lines that intersect. I would
suspect the number of such lines is about 10 and there are roughly
5-10 points for each line.
[/quote]
I assume you are speaking of _straight_ lines.

I think you could use the Hough transform, with distance and angle as
parameters, to separate the points belonging to each line and get a
first approximation of the parameters.

Then if you need more precision you should fit straight lines to the
subsets you got. See eg David Alciatore, Rick Miranda, The Best Least-
Squares Line Fit, in Graphics Gems V for a method which minimizes the
normal squared distances.

You may also want to use a statistical method like Least Median of
Squares to get rid of possible outliers.

AJ · Guest

On Jun 20, 4:22 pm, Thomas Richter <t...@math.tu-berlin.de> wrote:
[quote]AJ schrieb:

Hi All,

I was hoping someone would be able to explain why something I have
noticed during some image processing analysis occurs.

When I am working with extremely degraded images, and am trying
different techniques to improve them, I often measure the PSNR (Peak
Signal-to-Noise Ratio) of the degraded image against the original,
undamaged image. Often, I find that the techniques that give the best
(highest) PSNR dB value are NOT the ones that give the best subjective
quality i.e. I chose restoration techniques that give a lower PSNR
because they LOOK better.

Can anyone explain why this happens?

PSNR relates bad to human vision. You should try different metrics
with a better correlation to subjective quality.

Recommendations: MS-SSIM (Sheik & Bovik) works quite ok (though not great),
VDP (Daly>s Visual Difference Predictor) works much better, though it is
grey-scale only and it is extremely slow.

Maybe it is something to do with
the fact that the image quality of the degraded image is so poor to
start with, that using PSNR is not that helpful?

If you want to measure subjective quality, PSNR is not very helpful to begin with...

So long,
Thomas
[/quote]
Hi Thomas,

Many thanks for the reply.

I am using Matlab for my experimentation. Do you know if there is a
Matlab implementation of MS-SSIM (Sheik & Bovik) or VDP (Daly>s Visual
Difference Predictor)?
I have had a google but no luck. I would like to avoid re-writing any
code if it is already available, to save time and effort!

Kind regards,

AJ

Andrew_M · Guest

I was offered once to write similar prog, more complicated- there was
3D problem. I refused, but I>d thought about this prob, so I can
share some of my ideas. At the first, one needs to set at least one
point at the first image and corresponding point at the second. Split
the first image to set of small pieces. Each piece has its own buddy
at the second image. If relief is more or less flat, task is much
simpler. To transform one piece to its buddy one must set 6
parameters. Two of them obviously are transaction (may be shift is
better term) vector coordinates, and three others describe affine
distortion of a piece, caused by the fact, that points of view and
camera>s position aren>t the same. One may calculate parameters of
affine distortion of a single piece, using Fourier transform (it is
better than direct correlation method). When this op is complete for
the first piece, it have to be repeated for immediate neighboring of
he first piece. Before doing this, one have to forecast set of 6
parameters for each next piece. Its real parameters must be near to
recalculated set. This method, I think, will work if both of images
are similar (not a great difference in position and so on). Contact
me, if you need more detailed info

Thomas Richter · Guest

AJ schrieb:

[quote]Hi Thomas,

Many thanks for the reply.

I am using Matlab for my experimentation. Do you know if there is a
Matlab implementation of MS-SSIM (Sheik & Bovik) or VDP (Daly>s Visual
Difference Predictor)?
[/quote]
There is surely one for MS-MSSIM since S & B originally wrote this in
Matlab. Usually, you can download the code from their home page.

http://www.ece.uwaterloo.ca/~z70wang/research/ssim/

It is actually one of the first hits that come up - did you try google? (-:

Please check whether there>s a multiscale version available on the page
- the single-scale version doesn>t really perform very well IMHO.

There is AFAIK no matlab version of VDP, it>s C++ only (and there>s
nothing bad in using that instead of matlab. :-).

There was also a work based on DCT by Ponomarenko et al, using matlab,
unfortunately their web-page seems to be down:

http://www.cs.tut.fi/~ponom/psnrhvsm.htm

On Sheik/Bovik>s home page, you>ll also find the matlab source for
"VIF", another model. I haven>t used that yet, so I can not comment on
it, but feel free to try.

So long,

Thomas

Nasser Abbasi · Guest

"Nasser Abbasi" <nma@12000.org> wrote in message
news:a%x8k.9780$xZ.8316@nlpi070.nbdc.sbc.com...
[quote]hello;

I have 2 images (of same size), both gray levels. Say A and B. B is a
modified version of A, and I>d like to measure how far B is from A.

I currently use RMSE and also use relative error,absolute error and
histogram difference. But I 'like' RMSE the most so far.

Is there a better method to quantify, in one number, the difference
between 2 images that I am overlooking ?

thanks,
Nasser
[/quote]
Thanks to all who replied.

fyi, I also found 'peak SNR'. Which is, from wikipdia

"The PSNR is most commonly used as a measure of quality of reconstruction in
image compression etc. "

http://en.wikipedia.org/wiki/PSNR

Nasser

newsy tepsy · Guest

Thomas Richter pisze:
[quote]AJ schrieb:
Hi All,

I was hoping someone would be able to explain why something I have
noticed during some image processing analysis occurs.

When I am working with extremely degraded images, and am trying
different techniques to improve them, I often measure the PSNR (Peak
Signal-to-Noise Ratio) of the degraded image against the original,
undamaged image. Often, I find that the techniques that give the best
(highest) PSNR dB value are NOT the ones that give the best subjective
quality i.e. I chose restoration techniques that give a lower PSNR
because they LOOK better.

Can anyone explain why this happens?

PSNR relates bad to human vision. You should try different metrics
with a better correlation to subjective quality.

Recommendations: MS-SSIM (Sheik & Bovik) works quite ok (though not great),
VDP (Daly>s Visual Difference Predictor) works much better, though it is
grey-scale only and it is extremely slow.

Maybe it is something to do with
the fact that the image quality of the degraded image is so poor to
start with, that using PSNR is not that helpful?

If you want to measure subjective quality, PSNR is not very helpful to begin with...

So long,
Thomas
i>d also recommend to have look at the S-CIELab[/quote]
(http://white.stanford.edu/~brian/scielab/)
It is:
- implemented in Matlab
- reasonably fast (or slow)
- well documented

You can use it to compute weighted image quality/distortion in both ways
or WMSE or WSNR.

It computes color difference as CIE deltaE distance (Euclidean in Lab
space, prefiltered with eye CSF functions).

regards
Przemek

P.S
There is also minor bug in computing visual angle but its not critical
and result differences are small. If someone is interested i can send a
"patch".

Przemys?aw Skurowski · Guest

Thomas Richter pisze:
[quote]AJ schrieb:
Hi All,

I was hoping someone would be able to explain why something I have
noticed during some image processing analysis occurs.

When I am working with extremely degraded images, and am trying
different techniques to improve them, I often measure the PSNR (Peak
Signal-to-Noise Ratio) of the degraded image against the original,
undamaged image. Often, I find that the techniques that give the best
(highest) PSNR dB value are NOT the ones that give the best subjective
quality i.e. I chose restoration techniques that give a lower PSNR
because they LOOK better.

Can anyone explain why this happens?

PSNR relates bad to human vision. You should try different metrics
with a better correlation to subjective quality.

Recommendations: MS-SSIM (Sheik & Bovik) works quite ok (though not
great),
VDP (Daly>s Visual Difference Predictor) works much better, though it is
grey-scale only and it is extremely slow.

Maybe it is something to do with
the fact that the image quality of the degraded image is so poor to
start with, that using PSNR is not that helpful?

If you want to measure subjective quality, PSNR is not very helpful
to begin with...

So long,
Thomas
i>d also recommend to have look at the S-CIELab[/quote]
(http://white.stanford.edu/~brian/scielab/)
It is:
- implemented in Matlab
- reasonably fast (or slow)
- well documented

You can use it to compute weighted image quality/distortion in both ways
or WMSE or WSNR.

It computes color difference as CIE deltaE distance (Euclidean in Lab
space, prefiltered with eye CSF functions).

regards
Przemek

P.S
There is also minor bug in computing visual angle but its not critical
and result differences are small. If someone is interested i can send a
"patch".

lomas · Guest

On Sep 22, 3:39 pm, lijing <verylij...@163.com> wrote:
[quote]the image is gray level
first the face image is devided into small block,then extract the
histogram from these small image block.

now,how to design a weak classifier or what kind of measures can been
used to distinguish useful histogram feature from redundant
histogram feature? could you give me some help? Thanks!
[/quote]

common boosting algorithm perfers to haar-like features, because they
are cheap in computation(I guess).
due to histogram is 1D, you may try 1D haar-like features.

dondora · Guest

thanks for your help.
It helps a lol although I need some patience to learn. ^.^
oh do you have that one in c++ version?

Jonathan Campbell · Guest

dondora wrote:
[quote]thanks for your help.
It helps a lol although I need some patience to learn. ^.^
oh do you have that one in c++ version?
[/quote]
Unfortunately no. We developed a C version long ago but it has become
lost in the mists of time.

Provided you have an image processing API (*) (there>s the beginnings of
a C++ one mentioned on my website, but I>d advise seeking out a more
mature one), and a function to compute L = A^T A (either SVD or straight
eigen-decomposition) then conversion is trivial. If you start the job,
I>d be happy to help; see my email address on the website. But if you
are just learning programming, then the task is less trivial and I don>t
want to teach C++ *and* program an application.

Note that my Java application is written in a linear and naive manner
--- i.e. follows the Turk-Pentland paper almost exactly.

(*). If you attempt to work with raw arrays, i.e. interspersing your
*application* (face recognition) with your *image processing* software,
then the task will be very difficult and more error prone.

Incidentally, if you want to learn properly the mathematics behind all
this, including fabulous videos of Prof. Strang lecturing, see:

http://web.mit.edu/18.06/www/

I wish I could suggest a small toy data set (with test inputs and
intermediate and final outputs) upon which you could test your program,
but I cannot. You might be able to find the images used in the paper; at
least, you will find a similar data set.

Best regards,

Jon C.

Jonathan Campbell · Guest

dondora wrote:
[quote]thanks for your help.
It helps a lol although I need some patience to learn. ^.^
oh do you have that one in c++ version?
[/quote]
Unfortunately no. We developed a C version long ago but it has become
lost in the mists of time.

Provided you have an image processing API (*) (there>s the beginnings of
a C++ one mentioned on my website, but I>d advise seeking out a more
mature one), and a function to compute L = A^T A (either SVD or straight
eigen-decomposition) then conversion is trivial. If you start the job,
I>d be happy to help; see my email address on the website. But if you
are just learning programming, then the task is less trivial and I don>t
want to teach C++ *and* program an application.

Note that my Java application is written in a linear and naive manner
--- i.e. follows the Turk-Pentland paper almost exactly.

(*). If you attempt to work with raw arrays, i.e. interspersing your
*application* (face recognition) with your *image processing* software,
then the task will be very difficult and more error prone.

Incidentally, if you want to learn properly the mathematics behind all
this, including fabulous videos of Prof. Strang lecturing, see:

http://web.mit.edu/18.06/www/

I wish I could suggest a small toy data set (with test inputs and
intermediate and final outputs) upon which you could test your program,
but I cannot. You might be able to find the images used in the paper; at
least, you will find a similar data set.

Best regards,

Jon C.

P.S. If you post an accessible version of the OpenCV code link, I>ll
have a look at it. j.c.

dondora · Guest

I didn>t see all of your post all. first of all I think it is good to
write a link to the source code.
here it is.

http://mfiles.naver.net/6dbe598296c7a0163f58/data35/2008/10/6/247/pca-powerqn.zip

I>ll see your post at home reply on it.

Jonathan Campbell · Guest

Jonathan Campbell wrote:
[quote]dondora wrote:
thanks for your help.
It helps a lol although I need some patience to learn. ^.^
oh do you have that one in c++ version?

Unfortunately no. We developed a C version long ago but it has become
lost in the mists of time.

Provided you have an image processing API (*) (there>s the beginnings of
a C++ one mentioned on my website, but I>d advise seeking out a more
mature one), and a function to compute L = A^T A (either SVD or straight
eigen-decomposition) then conversion is trivial. If you start the job,
I>d be happy to help; see my email address on the website. But if you
are just learning programming, then the task is less trivial and I don>t
want to teach C++ *and* program an application.
[/quote]
If you must use C++, it seems to me that OpenCV may be the API/toolkit
you need, though it is written in C; C code will be callable from C++.

There are examples at:

http://www.cognotics.com/opencv/servo/index.html

At least on Linux, OpenCV is very easy to install and use.

I still cannot access the link you give, but don>t trouble to fix it; I
have enough examples to do.

Best regards,

Jon C.


www.GetXFactor.com Leading Technology, Science, Agriculture News and information	Part of the Identityscape.com network... getxfactor.com jmoodmusic.com smartbusinesschoices.com mintdepot.com lowfaresalways.com evangelicalview.com shoppingpodder.com soproudlywehail.com webnews.ws currenthumor.com

Science and Technology news... Forum Index -> Image Processing Forum	Goto page 1, 2 Next
Page 1 of 2