# f-number equivalent between m43 and FF

Started Mar 25, 2014 | Discussions thread
Re: Noise is good ?

crames wrote:

Detail Man wrote:

Aside from the (in practice unusual) special cases of resampling (up or down) by odd integer factors (3, 5, 7, ... etc.), resampling operations (as I understand them) involve a first step of interpolation (upsampling) by the integer value of the numerator of the (net, composite) resampling ratio, followed by low-pass filtering, followed by a second step of decimation (downsampling) by the integer value of the denominator of the (net, composite) resampling ratio.

Have in some cases purposely downsampled processed images for presentation by odd integer factors (3, 5, 7, ... etc.) in the hopes of generating less "numerically massaged" results. At times (particularly in detailed landscape images with a lot of relatively high spatial frequency content) this (may, perhaps) result in slightly (subjectively perceived as) "better" looking results.

What is the problem with odd integer factors?

None. The idea of collapsing/expanding odd numbered NxN pixel array data into a single pixel (or expanding single pixel data into odd numbered NxN pixel arrays) seems potentially a good one.

Specific algorithms or implementations have problems with them?

There are a lot of ways to "zoom" an image, including by fractional factors.

My intended point was that fractional factors (involving interpolation followed by low-pass filtering followed by decimation) are (when) resizing to a given physical display size unavoidable.

Note: I was not reading your expressed thoughts as indicating that non-fractional was a necessity.

Here is a case where I downsampled by an integer factor of 3 (16-bit arithmetic, Lanczos-3):

Pixel dimensions = 1345 x 1009

However, I have invariably found myself desirous of "filling up" my 1200 pixel-height monitor screen, and have later opted for leaving my "numerology" behind - in favor of obtaining the larger physical viewing size made possible by filling up my monitor display-screen on each occasion.

Close-up inspection of any visible differences involves using (additional) upsampling algorithms applied by the viewer application utilized (the specific nature of which the user often has does not have specific knowledge of, or any control over) - which rather "muddies the waters" of analysis ?

The more high-frequency content (your example images go right up to Nyquist and a maybe little beyond) the more need for a high-quality upsampling. Nearest neighbor would be the worst but is usually what programs use for quick display.

Lanczos is one of the best but not always available.

Perhaps that use of "NN" in viewing apps itself is the origin of expressed "blotchiness" perceptions ?

As you have more recently stated:

Here is a comparison with nearest neighbor, which shows "pixelation" that is solely a result of the nearest neighbor upsampling and is not present at all in the original image.

This downsampled version (using the same 16-bit arithmetic, Lanczos-3) had the same mild (16-bit) USM applied utilizing the same parameters (Radius=0.33 pixel, Strength, and Threshold):

Pixel dimensions = 1600x1200

I am not suggesting that this be part of a workflow (most images are purposedly not enlarged to the point that a displayed pixel is larger than human CoC when projected on the retina, otherwise posterization becomes visible and objectionable). I am simply saying that to compare a visibly posterized image to one that is not is unfair to one and not apples-to-apples. In that case adding noise is one way to level the playing field.

GB's argument has to do with photon counts and photon and read noise. These should all be measurable.

Visibility of noise is a completely different can of worms that IMO is best kept out of the discussion or made into a separate one.

Makes sense. Maintaining the same physical viewing size (paradoxically, it seems) necessarily involves the use of additional display resampling operations when engaging in making visual comparisons.

The problem with comparisons is that "current DPR dogma" requires that they be made at the same size, ...

Which seems to me to be an understandable desire when comparing images (or image portions).

... usually meaning that the images being compared do not get equal treatment/processing, which further muddies the waters.

As you have more recently stated:

The problem is that the image with more pixels will (almost always) have the advantage. Anything you do the equalize the display sizes will degrade/disadvantage one or the other, or both.

So it (might) seem that - given the use of a "suitable" resampling algorithm - the decision to downsample one image, or to upsample the other image when making comparisons that normalize the pixel-dimensions of compared images (or portions thereof), is (potentially) an arbitrary one ?

This, however, leaves us with the seemingly endless myriad of subtleties surrounding the particular choice of multiplying windows utilized in reconstruction kernels explored and described in the related literature (where the particular characteristics of the processed image-data itself, as well as the matter of whether either upsampling or downsampling is being performed) are reportedly specifically relevant - as folks such as Nicolas Robidoux have explored in quite extraordinary detail (with wisely stated qualifications regarding perceptual subjectivities).

Thus, while bobn2's stated (perceptual) condition:

... 'viewed the same size' and the implicit condition is 'such that pixellation isn't visible in the viewed image' - that is, that the sampling rate in the viewed image is high enough to present an apparently analog image.

... is a readily understandable one, it seems that we are in practice not fundamentally able to transcend the "perceptual prisms" through which we experience the world - except where it comes to cases where the (internal) "filters" arising out of the limitations of our own perceptual facilities render both items of comparison as essentially equally "compromised" (in terms of visual resolution).

These limitations seem to me to be a situation of the sort not unlike my previous statement:

Perhaps the metaphor "signal" relates primarily to what we expect to be, and/or are pleased by, perceiving - and the metaphor "noise" arises as conceptualizations regarding any various remainders.

As is also the case in matters surrounding what particular stimuli may please our subjective aural perceptions (in terms of cordance and discordance), the "mind's eyes" of individual beholders render the possibilities of declarations of consensus of subjective preferences and tastes rising to some sort of established experiential objectivity as (while perhaps comforting) ... arguably dubious.

Complain
Post ()
Keyboard shortcuts: