[cbaytan]

IMO all the hassle about MP number in here is all about pixel density/diffraction, info is below. the answer is yes but only superb glass can handle more MP's on a APS-c size sensor, and thats why 12MP D700 has better IQ than 15MP K-7.

Diffraction Limited Photography: Pixel Size, Aperture and Airy Disks

[falconeye]

Isn't it a fact that regardless of the number of megapixels, the captor's photosites SIZE do matter?

As I am tired to explain, I really wish you read before you post, esp.

DxOMark - More pixels offset noise!

I do endorse the above article.

[mattdm]

Yes and of course when you upgrade a camera it's always a good idea to get the sensor to perform better or at least on the same level as the previous model.... Oh wait, they managed to bring out an upgrade that actually degrades image quality when it comes to dynamic range, noise performance and color depth: Compare cameras

Numbers, numbers, numbers — not the whole story. From all I've seen, the K-7 is much less susceptible to patterned noise than the older version of the sensor. Honestly, that's worth a little more noise overall.

[dlacouture]

From all I've seen, the K-7 is much less susceptible to patterned noise than the older version of the sensor. Honestly, that's worth a little more noise overall.

I agree! My K7 produces much more usable high-iso pics than my K20 ever could, as it does not suffer from patterned noise and magenta edges...
Anyway, I now also have a Kx for those it's-so-dark-I-cannot-even-focus shots...

[falconeye]

Numbers, numbers, numbers — not the whole story.

If testing sites would know their job it would be. After all, pattern noise is ... noise

[Gimbal]

If testing sites would know their job it would be. After all, pattern noise is ... noise

Yes, but it is an interesting point. How do you measure and value pattern noise contra non pattern noise?

[falconeye]

Yes, but it is an interesting point. How do you measure and value pattern noise contra non pattern noise?

I am 50% thru the development of a new image quality testing method for LumoLabs. It would actually measure pattern noise.

The method basically uses photographs of a fractal test pattern (where I would have to patent or at least copyright the exact pattern) and the spatial power spectrum of original pattern and its photos. The difference is the defect over spatial frequency, after proper treatment of lens and illumination artifacts. Noise would be part of this defect curve.

The great thing is that this measures a camera performance independently from technology. A higher pixel count leads to better performance at higher frequencies but not more noise at any given frequency. Fixed pattern noise shows up as peaks in the defect curve at lower spatial frequencies. A bad dynamic range shows up at lower spatial frequencies as well. And analog film cameras / film can be measured in the exact same way!

[Digitalis]

I am 50% thru the development of a new image quality testing method for LumoLabs. It would actually measure pattern noise.

The method basically uses photographs of a fractal test pattern (where I would have to patent or at least copyright the exact pattern) and the spatial power spectrum of original pattern and its photos. The difference is the defect over spatial frequency, after proper treatment of lens and illumination artifacts. Noise would be part of this defect curve.

The great thing is that this measures a camera performance independently from technology. A higher pixel count leads to better performance at higher frequencies but not more noise at any given frequency. Fixed pattern noise shows up as peaks in the defect curve at lower spatial frequencies. A bad dynamic range shows up at lower spatial frequencies as well. And analog film cameras / film can be measured in the exact same way!

sounds very interesting Falconeye...I have always thought there should be a way of using fractals as a way of measuring performance of optical imaging systems. though you would need an interesting fractal pattern to test the entire imaging field the Mandelbrot set isn't well suited to such tests. I have used the lichtenberg figure to experimentally test some of my lenses, though field curvature made such testing difficult.

[falconeye]

I have used the lichtenberg figure to experimentally test some of my lenses

A, a brother in soul

The Lichtenberg figure may have a two small fractional dimension. It's almost 1 dimensional, like the Mandelbrot set's border.

A started to think in terms of generating functions as well. But now, I think a generalization of the simple idea behind the Cantor set is a better approach. Unlike Mandelbrot an alike, the Cantor set can be easily redefined for 2D.

I am still undecided about the exact interval function (shouldn't be a step function really) and how to treat colors.

It may be less suitable to test lenses because lens properties aren't location-independent. But for sensor tests using a "best" lens in its sweet spot it should work out ok. Unevitable non-uniformity of lighting is another challenge. But it should show up in lowest spatial frequencies only. Where it could be measured for compensation.

[dlacouture]

Unevitable non-uniformity of lighting is another challenge. But it should show up in lowest spatial frequencies only. Where it could be measured for compensation.

Well, maybe a ring flash, or an uniformly back-lighted target, would take care of this?

EDIT: whoops, went right into dumb mode about the ring flash, as the inverse square law still applies... So, this leaves the back-lighted target...

[Digitalis]

Well, maybe a ring flash, or an uniformly back-lighted target, would take care of this?

EDIT: whoops, went right into dumb mode about the ring flash, as the inverse square law still applies... So, this leaves the back-lighted target...

unless he went through the expense of getting a studio ringflash. It would be possible, they are very powerful and some are capable of providing a very high speed flash. The Elinchrom ringflash is rated at 1/2600th at full power and can accommodate a wide range of lenses. I'll have to get the t= figures but they are bound to be pretty quick knowing Elinchrom.

[Class A]

As I am tired to explain, I really wish you read before you post, esp.

DxOMark - More pixels offset noise!

I've referenced that article before. I'm not sure why it doesn't help more than it should.

[Class A]

I am still undecided about the exact interval function (shouldn't be a step function really) and how to treat colors.

Have you considered using something like the Peano curve or the Hilbert curve and use different colours for various iteration depths of the curve?

[falconeye]

Have you considered using something like the Peano curve or the Hilbert curve and use different colours for various iteration depths of the curve?

Space filling curves won't do it, I guess. You either print finite versions with a peak in frequency space or you get a uniform surface.

My current intuition uses the Cantor set as a starting point. Besides the Cantor dust (which isn't dense enough), the Sierpinski carpet is a well known generalization to 2D:


It has a Haussdorf dimension of 1.8928 and any fractal around 1.7 to 1.9 seems suitable. List of fractals by Hausdorff dimension - Wikipedia, the free encyclopedia lists many of them in a convenient way.

However, I would alter the generation rule. Generalized Cantor sets use a rule to use a uniform plane, tile it into congruent parts, and remove some of them.

I think to alter this rule: I replace a single tile center by N centers of congruent parts. Then, I "attach" a fast decaying function to the centers like a Gaussian with a standard deviation proportional to the current iteration's tile scale. For each tile center, I use a color multiplication factor where black corresponds to the removal of the subtile. Then, I overlay all tile functions with the image of the previous iteration, where the choice of the exact overlay function (e.g., a multiplication) defines much of the final fractal. The choice of Gaussians should make the power spectrum smooth.

That gives me quite some design space and I'll have to write a program to navigate thru the possibilities.

Nevertheless, I'll need hard borders to measure a camera's capability to depict line structures. The above construction only measures a camera's capability to depict point structures. So, I'll have to combine several approaches. So, a standard duochrome fractal probably won't do the job.

[Inspector 17]

yes, numbers everywhere...

I was excited when the K7 was introduced and at first glance decided I would buy one when the price came down..

UNTIL I noticed that there were VERY VERY FEW examples of really good top class portraiture with the K7.

And while this is an observation rather than an exact quantification, the truth of the K7 is that there is a reddish cast to skin color and some sort of a frosted like muddiness because of this cast to my eyes.

Check Flickr, other example sites - there are thousands upon thousands of absolutely great portraits for the K10D, the K20D, even the older ist*'s, but a dearth of great people/portrait shots with the K7. I have just never liked people examples with it despite liking all the performance features that the K7 introduced.

And so I wait now ready to pay first adopter price for a worthy K7 replacement AND ESPECIALLY HOPEFULLY my first choice, a home run Pentax FF (my 31 and 77 Limiteds NEED a great FF!). I need the FF sized viewfinder to do what I want. The K10D's focus faculty and lack of usefulness are making me nuts

I love my Limiteds and at lower ISO the K10D can be stellar, but Pentax I need a home run FF AND a home run K7 replacement