Image compression

[John T. Sharp]

I am starting a project measuring features of hand and foot x-rays on 400 to 600 sets ( 4 each set) so it is a rather large project. I'm estimating 250-350 hours.  There is some controversy among colleagues who we also be participating as to wheter measurements are likely to be affected by transmitting the images in compressed TIFF format.  I will be using the Microsoft Photo Editor to decompress them and then making the measurements with an ImageJ plugin.  

Would anyone with the engineering/math/computer science background sufficiently broad to know whether this "lossless" compression-decompression system will influence the measurements comment on this problem.

John T. Sharp

[Yong Zhang-2-3]

Hi, John,

If the image compression is "lossless", then the measurements will not be
affected since no information is lost during the compression. The lossless
compression is a good way to store/transfer large dataset without any loss
of the useful information. But you have to make sure that the compression
is really "lossless" when using any software, because one format can be
either lossless or lossy. For example, JPEG format can be lossless or
lossy. So you must select lossless mode when saving your original image
into another format. If your software does not provide lossless/lossy
selection when saving images, then you'd better switch to another software
which has the selection.

I hope it helps.

Best,


Yong Zhang



John T. Sharp said:

> I am starting a project measuring features of hand and foot x-rays on 400
> to 600 sets ( 4 each set) so it is a rather large project. I'm estimating
> 250-350 hours.  There is some controversy among colleagues who we also be
> participating as to wheter measurements are likely to be affected by
> transmitting the images in compressed TIFF format.  I will be using the
> Microsoft Photo Editor to decompress them and then making the measurements
> with an ImageJ plugin.
>
> Would anyone with the engineering/math/computer science background
> sufficiently broad to know whether this "lossless"
> compression-decompression system will influence the measurements comment
> on this problem.
>
> John T. Sharp
>

[Aryeh Weiss]

Yong Zhang wrote:

> Hi, John,
>
> If the image compression is "lossless", then the measurements will not be
> affected since no information is lost during the compression. The lossless
> compression is a good way to store/transfer large dataset without any loss
> of the useful information. But you have to make sure that the compression
> is really "lossless" when using any software, because one format can be
> either lossless or lossy. For example, JPEG format can be lossless or
> lossy. So you must select lossless mode when saving your original image
> into another format. If your software does not provide lossless/lossy
> selection when saving images, then you'd better switch to another software
> which has the selection.
>

That's the good news. The bad news is that you may only see up to 50% size
reduction with lossless compression (unless your images are thresholded or have
large uniform areas), but if you cannot tolerate any discrepencies in you
images, then that's what you get.

If you plan to filter out point noise (eg, with a median filter), or to do any
smoothing, then you should do this before lossless compression. This will
improve the compression ratio.

--aryeh
--
Aryeh Weiss
School of Engineering
Bar Ilan University
Ramat Gan 52900 Israel

Ph:  972-3-5317638
FAX: [phone]

[Jeff Brandenburg]

Just now catching up on holiday email...

On Dec 22, 2005, at 8:33 PM, John T. Sharp wrote:

> I am starting a project measuring features of hand and foot x-rays on  
> 400 to 600 sets ( 4 each set) so it is a rather large project. I'm  
> estimating 250-350 hours.  There is some controversy among colleagues  
> who we also be participating as to wheter measurements are likely to  
> be affected by transmitting the images in compressed TIFF format.  I  
> will be using the Microsoft Photo Editor to decompress them and then  
> making the measurements with an ImageJ plugin.
>
> Would anyone with the engineering/math/computer science background  
> sufficiently broad to know whether this "lossless"  
> compression-decompression system will influence the measurements  
> comment on this problem.

As others replied, "lossless" means lossless -- compressing losslessly,  
then decompressing, should result in bit-for-bit identical files.  My  
limited Windows experience leads me to distrust any claims of  
"lossless" image treatment in any Microsoft application, but I have no  
evidence that Photo Editor does the wrong thing.

While it isn't directly relevant to the problem you describe, I  
recently found this press release indicating that lossy JPEG  
compression of mammography images can actually IMPROVE (!) diagnostic  
accuracy:

http://news.uns.purdue.edu/UNS/html4ever/2005/ 
051220.Lucier.mammograms.html

If the standard JPEG compression algorithm can do this, one wonders  
what specialized algorithms might be able to achieve...
--
        -jeffB (Jeff Brandenburg, Duke Center for In-Vivo Microscopy)

[Gabriel Landini]

On Thursday 05 January 2006 20:27, Jeff Brandenburg wrote:
> recently found this press release indicating that lossy JPEG
> compression of mammography images can actually IMPROVE (!) diagnostic
> accuracy:
>
> http://news.uns.purdue.edu/UNS/html4ever/2005/
> 051220.Lucier.mammograms.html

Thanks for the link, Jeff.
Strange claim, isn't it?
However, what is reported there does not seem to be quantitative analysis of
images.
Instead it seems that the author reports on subjective visual assessment by
radiologist observers. That is a completely different issue. (specially
because there is a degree of subjectivity in visual perception).

It may be that the compression artifacts do not interfere with perception of
the features of interest, or (as hinted) that it removes some noise. But I
think that it is a very inefficient and unpredictable way of performing noise
reduction.

Still, I think that for quantitative image analysis, it is better to start
with loss-less compressed images. If necessary noise reduction and image
simplification can be achieved afterwards with more controlled methods than
the jpeg compression.

Cheers,

Gabriel