quality metric PSNR??

Started by rukmanimohan04 April 12, 2005

hi,

I am a bit confused regarding the use of PSNR... Is it absolutely a
blunder to compare an original image and its 2D DCT compressed-
decompressed version using this metric PSNR?? Can any one give me some
appropriate values for PSNR in this case? I want to know if this kind
of comparison based on PSNR value is really outdted????
waiting for ur reply,
regds,
rukmani




Hi Rukmani

PSNR is a standard way of measuring the quality of
compressed images. It is absolutely not outdated! Even
papers on JPEG 2000, if you have seen them, compare
baseline JPEG and JPEG 2000 on the basis of PSNR vs
bits/pixel values. (well, JPEG 2000 isnt recent
anymore, but you get the idea)

PSNR = 10*log10 (255^2/MSE)

where original image is assumed to be use 8 bits per
pixel (hence 'peak' = 255).

Usually PSNR of more than 35 dB is considered good
quality.

If you are dealing with colored images, you would work
with the Y,U,V components separately and quote either
the average PSNR in dB or the three PSNR values
separately (psnrY, psnrU, psnrV) to indicate the
quality of the compressed image.

hope this helps to some extent.
Sameer

----
"I was born not knowing and have had only a little
time to change that here and there." - Richard Feynman --- rukmanimohan04 <rukmanimohan04@rukm...> wrote:
>
> hi,
>
> I am a bit confused regarding the use of PSNR... Is
> it absolutely a
> blunder to compare an original image and its 2D DCT
> compressed-
> decompressed version using this metric PSNR?? Can
> any one give me some
> appropriate values for PSNR in this case? I want to
> know if this kind
> of comparison based on PSNR value is really
> outdted????
> waiting for ur reply,
> regds,
> rukmani >
>


________________________________________________________________________
Yahoo! India Matrimony: Find your life partner online
Go to: http://yahoo.shaadi.com/india-matrimony



Hi Rukmani, I feel PSNR and MSE metrics are good for general quality assessed by the human eye. However a computer will feel different. If you need to detect or segment an object or classify the image PSNR tells you nothing. If say blurring artifacts or edge artifacts are introduced you are hosed. I suggest you evaluate what the end user will be doing with the imagery and apply an appropriate evaluation such as classification rate, etc. However for applications such as medical diagnosis the end user is using their vision and PSNR can be sufficient unless the doctor is using edges or other features as clues. Then you may want to compare edge maps or texture features and compare the values pre/post compression. Good Luck, Mark > > > > >hi, > >I am a bit confused regarding the use of PSNR... Is it absolutely a >blunder to compare an original image and its 2D DCT compressed- >decompressed version using this metric PSNR?? Can any one give me some >appropriate values for PSNR in this case? I want to know if this kind >of comparison based on PSNR value is really outdted???? >waiting for ur reply, >regds, >rukmani > > > > > > > > >
Hi Rukmani, I feel PSNR and MSE metrics are good for general quality assessed by the human eye. However a computer will feel different. If you need to detect or segment an object or classify the image PSNR tells you nothing. If say blurring artifacts or edge artifacts are introduced you are hosed. I suggest you evaluate what the end user will be doing with the imagery and apply an appropriate evaluation such as classification rate, etc. However for applications such as medical diagnosis the end user is using their vision and PSNR can be sufficient unless the doctor is using edges or other features as clues. Then you may want to compare edge maps or texture features and compare the values pre/post compression. Good Luck, Mark > > > > >hi, > >I am a bit confused regarding the use of PSNR... Is it absolutely a >blunder to compare an original image and its 2D DCT compressed- >decompressed version using this metric PSNR?? Can any one give me some >appropriate values for PSNR in this case? I want to know if this kind >of comparison based on PSNR value is really outdted???? >waiting for ur reply, >regds, >rukmani > > > > > > > > >

Hi Rukmani

PSNR is a standard way of measuring the quality of
compressed images. It is absolutely not outdated! Even
papers on JPEG 2000, if you have seen them, compare
baseline JPEG and JPEG 2000 on the basis of PSNR vs
bits/pixel values. (well, JPEG 2000 isnt recent
anymore, but you get the idea)

PSNR = 10*log10 (255^2/MSE)

where original image is assumed to be use 8 bits per
pixel (hence 'peak' = 255).

Usually PSNR of more than 35 dB is considered good
quality.

If you are dealing with colored images, you would work
with the Y,U,V components separately and quote either
the average PSNR in dB or the three PSNR values
separately (psnrY, psnrU, psnrV) to indicate the
quality of the compressed image.

hope this helps to some extent.
Sameer

----
"I was born not knowing and have had only a little
time to change that here and there." - Richard Feynman --- rukmanimohan04 <rukmanimohan04@rukm...> wrote:
>
> hi,
>
> I am a bit confused regarding the use of PSNR... Is
> it absolutely a
> blunder to compare an original image and its 2D DCT
> compressed-
> decompressed version using this metric PSNR?? Can
> any one give me some
> appropriate values for PSNR in this case? I want to
> know if this kind
> of comparison based on PSNR value is really
> outdted????
> waiting for ur reply,
> regds,
> rukmani >
>


________________________________________________________________________
Yahoo! India Matrimony: Find your life partner online
Go to: http://yahoo.shaadi.com/india-matrimony




hi,

I am a bit confused regarding the use of PSNR... Is it absolutely a
blunder to compare an original image and its 2D DCT compressed-
decompressed version using this metric PSNR?? Can any one give me some
appropriate values for PSNR in this case? I want to know if this kind
of comparison based on PSNR value is really outdted????
waiting for ur reply,
regds,
rukmani