Image quality measure without having the original image?

Hello everybody,

I need a resource or information on how to measure the quality of a reconstructed image without having the original image.

Here is the set up(wireless multimedia sensors):
I have an algorithm to compress the image(encoder, image sensors) and send it to the base station for processing. But after decoding i need to measure the quality of the reconstructed image and give feedback to the encoder(in sensors) to send more refinment bits if it is below a certain threshold.

any information or resource on that?

Thank you, Start a New Thread

Sami-

> I need a resource or information on how to measure the quality
> of a reconstructed image without having the original
> image.
>
> Here is the set up(wireless multimedia sensors):
> I have an algorithm to compress the image(encoder, image
> sensors) and send it to the base station for processing. But
> after decoding i need to measure the quality of the
> reconstructed image and give feedback to the encoder(in sensors)
> to send more refinment bits if it is below a certain threshold.
>
> any information or resource on that?

Can you make assumptions about the compression type? For example if it's JPEG, then you could search the image for
edges and apply some measure for sharpness.

Another idea is whether you can make assumptions about what type of content to expect. People? Nature scenes? Cars?
If so then maybe you can look for histograms that seem reasonable, or even use some type of image recognition to
figure out whether an image "makes sense" or not.

But in any case your problem description sounds to me too casual; maybe you have to think more carefully about this.
If you can't make any assumptions, then you're basically saying "I want to measure the quality of any arbitary image".
In that case all you could do is apply generic measures for smoothness, randomness, noise, etc.

-Jeff Start a New Thread

Since the sensor has the computational resources to compress the image and
also has access to the uncompressed image, it would seem the best place to
measure the quality of the compressed image would be at the sensor rather
than at the decoder.

On Fri, Apr 16, 2010 at 4:57 PM, Jeff Brower wrote:

> Sami-
> > I need a resource or information on how to measure the quality
> > of a reconstructed image without having the original
> > image.
> >
> > Here is the set up(wireless multimedia sensors):
> > I have an algorithm to compress the image(encoder, image
> > sensors) and send it to the base station for processing. But
> > after decoding i need to measure the quality of the
> > reconstructed image and give feedback to the encoder(in sensors)
> > to send more refinment bits if it is below a certain threshold.
> >
> > any information or resource on that?
>
> Can you make assumptions about the compression type? For example if it's
> JPEG, then you could search the image for
> edges and apply some measure for sharpness.
>
> Another idea is whether you can make assumptions about what type of content
> to expect. People? Nature scenes? Cars?
> If so then maybe you can look for histograms that seem reasonable, or even
> use some type of image recognition to
> figure out whether an image "makes sense" or not.
>
> But in any case your problem description sounds to me too casual; maybe you
> have to think more carefully about this.
> If you can't make any assumptions, then you're basically saying "I want to
> measure the quality of any arbitary image".
> In that case all you could do is apply generic measures for smoothness,
> randomness, noise, etc.
>
> -Jeff
>
>
> Start a New Thread

Thanks Jeff for the information and i will look at your suggestted methods but to be more clear,
I have multimedia sensors monitoring a certain area(surveillance), and they have more than one encoder type. Once they acquire the image they have to compress it using bckground subtraction method or DWT ... which encoder to be used by the sensor should be determined by the base station. Let's say the bas station recieves bitstream compressed using bachground subtraction and check the quality of the reconstructed image and if it's below a certain threshold the background must not be good(may be due to sudden illumination change), so till a good representative background is made they have to encode using DWT. But you know that's where i come in to this problem. The original image is on the sensor side, the reconstructed image is on the base station side ... and i was looking for a way to measure the quality of the reconstructed image without having the original image(Note that the sensors transmit the compressed bitstream wirelessly ... )

Hello everybody,
>
>I need a resource or information on how to measure the quality of a reconstructed image without having the original image.
>
>Here is the set up(wireless multimedia sensors):
>I have an algorithm to compress the image(encoder, image sensors) and send it to the base station for processing. But after decoding i need to measure the quality of the reconstructed image and give feedback to the encoder(in sensors) to send more refinment bits if it is below a certain threshold.
>
>any information or resource on that?
>
>Thank you, Start a New Thread

Thanks for your trial to help Patrick, but i want to do it in the base station side/ decoder side since the sensors are battery powered and computing a quality metric for each encoded frame will impose more computation that will drain the battery power.

Hello everybody,
>
>I need a resource or information on how to measure the quality of a reconstructed image without having the original image.
>
>Here is the set up(wireless multimedia sensors):
>I have an algorithm to compress the image(encoder, image sensors) and send it to the base station for processing. But after decoding i need to measure the quality of the reconstructed image and give feedback to the encoder(in sensors) to send more refinment bits if it is below a certain threshold.
>
>any information or resource on that?
>
>Thank you, Start a New Thread