Path: news.mathworks.com!newsfeed-00.mathworks.com!newsfeed2.dallas1.level3.net!news.level3.com!postnews.google.com!news1.google.com!npeer02.iad.highwinds-media.com!news.highwinds-media.com!feed-me.highwinds-media.com!post01.iad.highwinds-media.com!newsfe03.iad.POSTED!7564ea0f!not-for-mail
From: Walter Roberson <roberson@hushmail.com>
User-Agent: Thunderbird 2.0.0.16 (Windows/20080708)
MIME-Version: 1.0
Newsgroups: comp.soft-sys.matlab
Subject: Re: I need to help?
References: <gb0kj5$ov3$1@fred.mathworks.com> <e7ea5034-2104-4ba0-8820-104944d5d5a3@f36g2000hsa.googlegroups.com> <abc8ddcc-faaf-43ff-9a8b-f543a930b335@e39g2000hsf.googlegroups.com> <gb9o8t$8jc$1@fred.mathworks.com>
In-Reply-To: <gb9o8t$8jc$1@fred.mathworks.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Lines: 66
Message-ID: <qg%Bk.30209$rV4.22804@newsfe03.iad>
NNTP-Posting-Host: 24.79.146.116
X-Complaints-To: internet.abuse@sjrb.ca
X-Trace: newsfe03.iad 1222148246 24.79.146.116 (Tue, 23 Sep 2008 05:37:26 UTC)
NNTP-Posting-Date: Tue, 23 Sep 2008 05:37:26 UTC
Date: Tue, 23 Sep 2008 00:38:06 -0500
Xref: news.mathworks.com comp.soft-sys.matlab:491466


ABUTALA wrote:
>> Walter said:
>>> Just out of curiosity, how does your course reference material define "quality" ?

> Quality means  the different between images detaction

"Image detection" implies that something is doing the
image detection -- that there is an image detector, whether electro-
mechanical or human. But different humans and different electro- mechanical
detection systems are relatively better at detecting specific image aspects
than other specific detectors. There are some things that humans in general
are excellent at detecting, and there are other things that even relatively
simple electro- mechanical systems are better at than most humans. Though
of course people like me who are very near-sighted are not very good at
any kind of image detection without our glasses!

So for your situation, who is doing the image detection or what
are the characteristics of the electro-mechanical image detector
(and how is the software for it programmed) ?


I am, by the way, being *completely* serious about the possibility
that the image detection is being done by a human and that "quality"
might be based upon what the human perceives, rather than upon what
a program calculates. There is a classic paper that studied the
feasibility of using JPEG DCT compression in medical x-rays. The
traditional assumption was that lossy compression would be unacceptable
for something as important as medical x-rays (chest, I think they were),
that "surely" the lossy compression would introduce features that were
not really present, or remove features that were present. So the
paper authors put together a panel of expert x-ray image readers,
and presented them with pairs of images compressed (or not) to some degree,
and had the experts rate the quality of the images (the experts weren't
told which image was which.) Using this blind-comparison method, the
authors of the paper were able to determine a level of JPEG lossy compression
which the experts were unable to tell apart from the images that had not
had any lossy compression at all. The amount of JPEG DCT lossy compression that
was undetectable to the x-ray experts was much higher than one might expect
(but I do not recall the amount at the moment.)

So for the purposes of that study, "quality" was very much based upon
"what is (un-) detectable to a trained and experienced human expert".
For -that- study, mathematical statistics comparing images were completely
unimportant in determinations of "quality".

For that kind of "quality" measurement, the -only- way to proceed is to
show the images to a human and have the human rate the quality somehow.
Preferably multiple humans, and preferably in a random order, and definitely
with the human not being told which image is which (not even any feedback
after each image about whether they were "right" or "wrong" about which image
was which.)

But that kind of measurement of "quality" is not the only valid measure
of "quality": "quality" is whatever you define it as for a given situation.
"Quality" is a word similar to "better": you have to ask yourself
"Better for what purpose? Better as measured how?".

ImageAnalyst has indicated that he believes that PSNR is a bad measure
of "quality" for your purpose. Based upon my suspicions of what you
-should- really be trying to do, I believe that he is right that PSNR
is not good for your situation. On the other hand, because "quality"
is situational, without objective meaning, if you were to *define* "quality"
in terms of PSNR for the purposes of your assignment, then of course it
would be (tautologically) be the right thing to calculate. (But I don't
know if the "decibels" calculated by your program has any relationship to
PSNR; you would have to look up the definition of PSNR to determine that.)