Your reflex is a funnel!

In these weeks, thanks or due to the challenging SLR Sigma SD15, I’m brushing off the old notions of photography, dating back to the late 90s, on film SLR. I’m focusing in particular on the theme of sharpness, or rather the resolving power.

I then dived into the web, still a great resource, where in some personal blogs, and especially old forums of photography enthusiasts a bit ‘old school, I found the typical old discussions about the not importance of megapixels, the influence of the format and the related theme of the most demanding APS-C sensors, etc. In short, the usual discussions by photographers between the pro and the passionate fussy.

The sacred Grail of photography: definition.

And finally here is the object of my research, the theme of the definition of a photographic system, SLR or mirrorless that is, and of its scrupulous measurement.

The reality in which we live is analogical, composed of infinite details, definable with the symbol (infinite). In systems, it is usual to indicate in lpm, that is, the number of lines per millimeter that a photographic lens is able to distinguish.

So, reality is at infinite lpm, that is, it has infinite lines of definition, while every lens has its technical limits, we put around 100 lpm: it means that the projected (and normalized) image towards the machine will be defined at 100 lpm.

Lpm, or lines per millimeter: the number of pairs of lines that an optical or printing system is able to solve.

Your SLR is a funnel

I use this metaphor by now grinded by the trend of funnel-marketing to make the idea of how the different steps of light towards the sensor are critical … Another image could be that of a chain, where the weakest link defines the strength of the same.

The funnel of the definition, version with Bayer sensor.

In any case, here are the different steps involved:

1) Reality.

As anticipated, the reality to be photographed is formed by infinite lpm, potentially up to the size of the atoms.

2) The lenses.

The lens system of your lens is the first step that light must travel to become a photographic file.

Based on measurements taken and found in different sources, the reference mainstream photographic targets, the most desired, are the Canon L series, which define around 90–90 lpm.
(In reality, the quality gradually decreases towards the edges of the image, but for ease I will measure at 90 lpm, imagining to use a Canon 85mm from 90lpm).

So, already during the first critical passage, the definition is set to 90 lpm!

3) The anti-aliasing filter (Bayer sensors)

In front of the sensor the anti-aliasing filter (Lowpass) is positioned, which serves to slightly reduce the contrast of the incoming image.

Why this choice? It is a necessity due to the characteristics of the common sensors (Bayer), which suffer from the Moirè effect and other artifacts. (Photographing a TV on, or a shirt with fine lines you will notice the strange lines, or waves not existing to the eye).
The anti-aliasing filter also helps to lose some lpm, but it’s hard to measure exactly how much: by convention, 10lpm.

4) The sensor

Finally the sensor. Normally it will be a Bayer type sensor, with a single layer of pixels organized in a grid of diodes in 4×4 groups. The basic module includes 4 pixels, of which 2 pixel-diode are sensitive to green, to red, 1 to blue. The sensor “sees” reality through a greenish mosaic, which is corrected just after the shot.
Sensors also have their own definition, which can always be defined in LPM. The noble megapixels, advertised loudly by the marketing of the photographic industry, once they are reached a sufficient quantity, are secondary in the discourse of the definition.

The formula for finding the resolving power of a sensor is as follows
Horizontal MPX / mm of the horizontal / 2 sensor

Basically, we divide the number of pixels — we take the horizontal ones -, by the number of horizontal millimeters of the sensor, (obtaining the number of linear pixels per millimeter), finally dividing by two.

This last step is because to see the lines, it will take at least pairs of pixels, one to define the black line, the other for the empty space.

Applying the formula to my SLR I got a 64 lpm for the Sigma SD15 and 87 lpm for the EOS 1000d. (These are the two SLRs used in the Crespi d’Adda test).

5) Demosaicing (Bayer sensors)

There is one last step that is left out, even in the most in-depth forums: the demosaicization process, necessary for DSLR cameras from Canon, Nikon, Sony and so on.

The demosaicing process made simple

I do not forget that the excellent 87 lpm of my Canon 1000d are actually defined by pixel-diode, or sub-pixel, in practice columns of pixels (always measuring the horizontal definition) that are sensitive to a single color. These are not true full pixels (pixel-output).

In fact, just after the shot takes place the process of demosaicing, which through an algorithm creates the readable Bitmap for the human eye to which we are accustomed.

This file is the result of a first interpolation: the information from the diode pixels are reconstructed and reorganized by the algorithm, which mediates the tones, and invents new pixels.

From the point of view of the search for the definition, we lose the idea of a ‘pixel perfect’ file projected by the lens.

Taking the example of two rather oblique black lines, coming from the AA filter at 80 lpm, they are actually defined in color at 40 lpm — because the color information by pixel-location in the Bayer sensor arrives every 2 horizontal pixels, 2 vertically — . After demosaicing, the lines are also made up of invented colored pixels, a phenomenon that takes the name of ‘color alias’.

NOTE. I think that the money spent in R&D by giants like Canon and Nikon in improving the Bayer sensor algorithms and raising megapixels, over the years will reduce this gap in definition.

In the case of Foveon, or film

The funnel of the definition, version with the Foveon sensor.

In case the sensor is of the Foveon type, for example of the second generation Foveon x3 of the Sigma Sd15, the funnel is simpler:

The Foveon x3 sensor, simplifying

– the image could reach 90lpm from the lens (the new 85 series ART will solve 90 lpm? For now I have not found sources!)

– reach the sensor without passing through anti-alias filters, therefore always at 90lpm

– the sensor (old) of the SD15 will record it according to its limit of 64 lpm as found before.

You get a very sharp bitmap file, with a sharp pixel definition, as I discovered using just the “just discrete” 17–50 Sigma 2.8 hsm IS lens. (For example, power lines are defined over the sky by a single line of pixel).

Other examples

In the case of using an SD1 Merrill, the sensor would record at 102 lpm, even with a margin on a 90lpm ‘L series equivalent’ lens.

It is well known in photography that when a sensor with relatively few megapixels records a hyper-defined image from a good lens, this does not bring problems.

The opposite is true, that is, when a dense and high-MP sensor receives an image from a mediocre lens (therefore low lpm), in which case all the defects are highlighted!

A 22MP-bayer EOS 5D mkIII defines 80 lpm, apparently less than the 1000d (87 lpm), but this is due to the larger sensor size. Always thanks to the 35mm size, will receive more light during the filming and will then be advantaged in the next I will talk later, or the press, which takes place with a magnification.


Your camera, which reflex or mirrorless, works like a funnel: at each step the definition loses something that can not be recovered: so each step must be up to the point, in view of the final result.
It’s also useless to add single excellent and hyperperforming elements, such as an expensive Zeiss or Leica 150 lpm lens, if it is frustrated by an anti-aliasing filter and an old-fashioned bayer sensor with less than 10MP!

The Canon system for choice at 80–90 lpm?

Some forums discussed whether Canon could launch super-high-definition lenses like Zeiss and Leica — niche products over 100 lpm -.
Apart from the cost of purchase and production, it could be a non-redundant choice at the moment: the (mainstream) professional sector from the early 2000s uses the EOS digital system at 90 lpm and does not seem to ask for more.
Current professional SLRs work around that range (80–90 lpm) without too much loss.

Canon EOS system, Canon’s adv for the 25th anniversary

Sigma, a separate niche

The case of Sigma is particular, with the search for the yield of fine details and color ‘pasty’ and realistic, since the SD1, with the Foveon of the first generation. The Japanese company has lower numbers than Canon, and continues to develop the Foveon sensor, the only conceptually different from the others.

Advertising image of Sigma for the launch of SD1 Merrill

In the case of my Sigma SD15 tested (SLR of 2010 from 5MP), good for sharpness, with files engraved and defined up to a single pixel. Just look at the light, texture and hair cables on an x3f file to check.
I have some doubts about what happens in the case of prints over A4, or crop, where the result could be limited by the lack of megapixels.


The theme of printing and interpolation will be dealt with in the next article of this photographic series, with enlargement tests from the shots made in Crespi with the two SLRs SD15 and EOS 1000d.

Help minodesign, share this post!

Giovanni Minelli

Art Director & Cross media designer. Si occupa di progetti di pubblicità, graphic design e web design, con un focus su corporate image, UI e image manipulation. Sperimentatore delle nuove tecnologie ed insieme nostalgico della grafica anni '70 e dell'adv '80-90s; alla ricerca di una sintesi contemporanea delle tre.  Ma anche ex fumettista dilettante, blogger e venditore ebay.