Nefertiti's Family

<blockquote id="quote"><font size="2" face="Verdana, Arial, Helvetica" id="quote">quote:<hr height="1" noshade id="quote"><i>Originally posted by neilderosa</i>
<br />Another interesting outcome; the imperfections and blotches in the image are brought out into sharper relief now too.
<hr height="1" noshade id="quote"></blockquote id="quote"></font id="quote">
I think that's good. That can serve as a counterpoint to those who would say that we're "removing" or "minimizing" undesirable data. Actually, we're even enhancing the undesirable data.

rd

<blockquote id="quote"><font size="2" face="Verdana, Arial, Helvetica" id="quote">quote:<hr height="1" noshade id="quote"><i>Originally posted by rderosa</i>
<br />Why should this make it more lifelike? Is it a self-fulling type of thing?<hr height="1" noshade id="quote"></blockquote id="quote"></font id="quote">As long as you are using a process that takes no account of what is in the picture, but simply operates on pixels or "bins", there is practically no dnager of influencing the appearance through bias. If you had read me the definition of "equalize" in the abstract, without applying it to any particular image, I would have said "Yes, that would make the Mars imagery look more like the human eye would see it." And that is a content-independent statement.

In fact, I thought the combination of automated brightness and contrast adjustments were already doing that. But I see now that they only spread the end values and let everything else spread proportionally. This "equalize" thing is just what the doctor ordered. The old way, if there are a lot of pixels in a small range of brightness, the regular process ignores that and just works on the tail extremes, which then enhances contrast in the dominant range very little. This new "equalize" process tries to increse contrast where it is needed most -- specifically in the brightness range where most of the pixels are. And it does so without having any idea what the content of the image might be.

This mimics the human eye's ability to adjust its own range of brightness and contrast so as to bring out detail in whatever we are looking at.

<blockquote id="quote"><font size="2" face="Verdana, Arial, Helvetica" id="quote">quote:<hr height="1" noshade id="quote">Does it make it look more lifelike because it is lifelike? Or because spreading the grayscales accross the entire spectrum on a "sort of lifelike image" makes it look more lifelike?<hr height="1" noshade id="quote"></blockquote id="quote"></font id="quote">There is no denying a selection effect here. You probably tried lots of standard processing techniques and selected the one that made it most facelike. But I stand by my statement that I or anyone would have agreed beforehand that "equalize" is something desireable for any MGS image from Mars if the goal is to make the scene appear more like what we would see with our own eyes if we were there.

That isn't true for many of the processing techniques. For example, using contrast "curves" can be controlled in a way that emphasizes what we would like to emphasize and fade what we would like to fade. That would definitely be introducing bias to the appearance. But I don't see any content-related aspect to this "equalize" adjustment.

So IMO, the fact that it makes the image stronger should probably be taken as an indicator that the image really is stronger. And yes, I'd have drawn a parallel conclusion if it had made the image weaker. But I concede that it is less likely this discovery would have been made if it had operated that way. -|Tom|-

Rich,

You might want to run the histogram equalize program on R07 & R12, just to see if you can eke some more information out of them.

Neil

<blockquote id="quote"><font size="2" face="Verdana, Arial, Helvetica" id="quote">quote:<hr height="1" noshade id="quote"><i>Originally posted by tvanflandern</i>
<br />So IMO, the fact that it makes the image stronger should probably be taken as an indicator that the image really is stronger. And yes, I'd have drawn a parallel conclusion if it had made the image weaker. But I concede that it is less likely this discovery would have been made if it had operated that way. <hr height="1" noshade id="quote"></blockquote id="quote"></font id="quote">
That's really great to know, for a couple of reasons. One, it re-inforces that these images are really real, so to speak, and two, it clarifies in my mind how these techniques must have been arrived at. I would think it's a fairly safe bet that there have been papers written on the exact subject you just so clearly laid out, how the "Human eye's (have the) ability to adjust its own range of brightness and contrast so as to bring out detail in whatever we are looking at."

Once that happened, it's a fairly straightforward step, communicating that to a software development team, and "voila" we have a Histogram Equalize algorithm. So, in a sense, you've resolved my need for a "greater truth" in all of this. It looks more lifelike, or "facelike" as you said, because it is mimicking something lifelike in the production of that image. Namely, the human eye!


rd

For what it's worth, here are the storm troopers and facing woman equalized:



Emanuel

One other important point I'd like to make on the subject of these image processing techniques.

One has to be very careful when "stacking" the various techniques on top of each other. The software is not so good that the user can do anything and then expect a step to make it better.

(BTW this has nothing to do with Emanuel's post, it's just something I saw while doing my last post)

For instance, while I was playing with R07, on one of my attempts, I adjusted Contrast first, and then ran Equalize. The end result was just white dots on a black background, sort of like the dot key. When I went back and looked at the histogram after each step, I realized I had created a non-existent histgram after the first step. In other words, by not adjusting brightness accordingly, I had knocked the data out of the useful range. There were no built in protections to prevent me from doing that.

On the other hand, if you look at the histogram after each step, and make an attempt to center it in the useful range with brightness adjustment, then run equalize the result is fairly good.

As I suspected though, equalize does not necessarily always give one the best image. I've seen many instances of variations that looked better to me (on the poorer images) where I just did contrast and brightness adjustments. So, I don't think it's going to be a "magic bullet".

rd