A black-and-white digital file can be produced two ways: by capturing colour through a filtered sensor and then discarding the colour information, or by capturing luminance directly on a sensor that was never filtered for colour in the first place. The two paths look similar in the final frame, but they differ in how much light reaches the silicon and how much spatial detail survives. The honest comparison is concrete, and the cleanest example is Leica’s own pairing of two cameras built on identical silicon: the M9 (2009) and the M Monochrom, announced on 10 May 2012. Both use the same Kodak KAF-18500 CCD, 35.8 by 23.9 mm, 18 megapixels on 6.8 micron pixels. The only difference is that the Monochrom has the colour filter array scraped off.

The Bayer Array and What It Costs

Almost every colour sensor records colour through a colour filter array (CFA) patented by Bryce E. Bayer at Eastman Kodak: US Patent 3,971,065, filed 5 March 1975 and granted 20 July 1976. The pattern is a repeating two-by-two tile of one red, one blue and two green filters, so green occupies half of all photosites, red and blue a quarter each. Green is oversampled deliberately, because the human visual system derives most of its luminance, its sense of brightness and fine high-frequency detail, from green wavelengths.

Each photosite measures only one primary, the band its filter passes; the other two values are estimated from neighbours in a step called demosaicing. So a Bayer chip directly measures roughly one third of the colour information and interpolates the rest. That interpolation falls hardest on chroma resolution. Luminance suffers less because it tracks the densely sampled green channel, but it does suffer: one widely cited measurement puts the effective luminance resolution of a Bayer sensor at around 0.58 times its nominal pixel count.

How Demosaicing Blurs an Edge

The mechanism matters more than the slogan. A typical demosaic reconstructs the green channel first, because green is sampled at two of every four sites and so has the densest grid to interpolate across. Red and blue are then derived by holding the local red-to-green and blue-to-green ratios constant, filling in their missing three-quarters from the green-anchored estimate.

Consider a hard black-to-white edge falling across the sensor. That is a high-frequency luminance event, and luminance lives in every channel. But on the red and blue channels, three of every four samples at the edge are guesses, reconstructed by a kernel that necessarily averages across neighbouring sites that straddle the edge. Averaging across a transition is, by definition, blur. The interpolation cannot place a clean boundary where it has no measured sample, so fine luminance structure is softened. A filterless sensor has no such problem: every photosite measures the full luminance value at its own location, one photosite to one pixel, with nothing inferred.

The Filter You Cannot See: No Anti-Aliasing Layer

There is a second, often larger, contributor to the Monochrom’s per-pixel acuity, and it is easy to miss. Colour sensors carry an optical low-pass filter, the anti-aliasing (AA) filter, a deliberate blur placed in front of the sensor. Its job is to smear detail finer than the demosaic Nyquist limit so that fine repeating patterns do not produce coloured moiré once the CFA data is interpolated. That blur costs sharpness on every frame.

A monochrome sensor has no chroma to alias, so there is no coloured moiré to suppress, and the AA filter can be omitted entirely. The Monochrom therefore gains acuity twice over: no interpolation, and no optical low-pass layer. The trade is that luminance aliasing can still appear, so fine fabrics, distant railings and roof tiles may show monochrome moiré that a colour-plus-AA camera would have smoothed away.

The Marketing Figure Versus the Measured One

Leica’s own claim for the Monochrom is that it delivers images “100% sharper” than monochrome derived from a colour sensor of comparable megapixels, in other words roughly double. Treat that as the manufacturer’s figure, not an independent result.

The measured reality is more modest. Popular Photography’s lab testing resolved roughly 2675 lines per picture height for the M9 at ISO 80 against about 2800 lph for the Monochrom at ISO 160. That is a real gain, but it is a few per cent, not a doubling. The “100% sharper” line is best read as marketing shorthand for the combined effect of no demosaicing and no AA filter, an effect that is genuine and visible in micro-contrast and edge crispness, while falling well short of twice the resolved detail.

Why the Sensitivity Rises

Light absorbed by a filter never reaches the photodiode. Each Bayer photosite sees only its own pass-band, so a green site discards most of the red and blue arriving at it. Strip the CFA and every photosite collects across the whole visible spectrum, capturing more photons per site at a given exposure.

That shows up in the rated speed. The M9’s base ISO is 160 (pullable to 80), with its range topping out at 2,500; the Monochrom’s base is 320, running to 10,000. At the bottom, 320 divided by 160 is exactly one stop of extra base sensitivity. At the top, 2,500 to 10,000 is two stops of additional headroom. The noise improvement has a mechanism too: more photons per photosite means a larger signal sitting above a fixed read-noise floor, so the signal-to-noise ratio rises and shadows stay cleaner and tonally separated rather than mushing into noise.

Contrast Lives on the Lens Now

The trade is absolute: a filterless sensor records no colour and cannot be converted back. There is a physical reason this matters for tonal control. Panchromatic silver-halide emulsion, and bare silicon, are intrinsically more sensitive to blue and ultraviolet than the eye is, so an unfiltered blue sky renders too light and the clouds wash out. On film you fix that optically; on a monochrome digital sensor you must do exactly the same, because the file carries no colour for software to weight after the fact.

The tool is a contrast filter over the lens, working by absorption. A yellow filter passes yellow and the longer wavelengths, orange and red, while absorbing blue and violet; orange and red push further, cutting more blue and green so the sky darkens progressively and atmospheric haze, which is scattered short-wavelength light, drops out. Each filter exacts a factor in exposure: a Yellow 8 (K2) is factor 2, one stop; an Orange 16 sits stronger than yellow; a Red 25 is factor 8, three full stops, and produces the darkest sky and the hardest haze cut.

A worked example. Shoot Ilford HP5 Plus at its box speed of EI 400, frame a midday landscape, and meter the blue sky so it would otherwise fall around Zone VI. Put a Red 25 on the lens to drag that sky down toward Zone III or IV for a dramatic near-black rendering, then open up three full stops to pay the filter factor, dropping your working exposure to an effective EI 50. A monochrome digital body needs the identical filter on the lens to reach the same look, while a colour camera could have approximated it later by weighting the blue channel down in conversion. With no colour recorded, the decision is made at the moment of exposure, glass on the front, the same discipline the sky over a sheet of HP5 has always demanded.