LEICA

Alternative views on the Leica world by Erwin Puts

The new monochrom camera, introduced by Leica, is is a derivative of the main M camera (246 versus 240) and uses the same CMOS sensor with 24 Mp, but without the Bayer filter. The general consensus claims that the monochrom has superior resolution when compared to the sensors with Bayer filter. These claims are made without any direct comparison between different models and must therefore be classified as unscientific.
The basic issue of any digital camera is the demosaicing algorithm, that resides partly in the camera software and partly in the postprocessing software on the computer. The simple problem is this. The Bayer filter implies that every spatial location (pixel) on the sensor surface captures only one color component. This means that any pixel has a luminance value (brightness value) and a color value (R or G or B). A realistic color for any pixel has to be interpolated by looking at the neighboring pixels and using this information to calculate a color value that might be close to the original color when capturing the scene. There are may different algorithms that try to accomplish the most realistic reproduction. There are the bilinear, Cok, Freeman, LaRoche and many other algorithms. The main point is that every algorithm generates aliasing problems. A prefiltering of the image (a sightly blurring effect) may reduce the aliasing, but it will remain visible.
The Monochrom image is still composed of three layers (RGB), but the interpolation has an easy job because the color information does not need to be interpolated and therefore the aliasing is much less. This will improve the resolution of the image. But by how much?
To analyze this problem I made a simple test. I used the Apo-Summicon-M 75 mm at an aperture of 1:5.6 and the Siemens test chart by Image Engineering. The lighting was provided by the Metz flash unit and the cameras were fixed on a tripod. The cameras used were the Monochrom CCD, the M8.2 CCD and the M CMOS. The image files were processed by Iridient Developer all at the same parameters. The remarkable fact is that I had to enlarge the images to 400% to see significant differences.
Below is the image made with the M8.2. The finest details are blurred, but the general picture is crisp and clean.

M8-400%

Below is the image made with the M. The object is reproduced at a larger scale, because the file size is more than twice the number of pixels. There is however only a slight difference in quality! The thin filter of the M8 might be the cause of this phenomenon. It is also a proof that the performance of the M is often overhyped when compared to the original M8.

M240-400%

Below is the image made with the Monochrom/CCD.
MM-400%

Now we see excellent resolution that is better than what the M8 and M do deliver, because there is no color to interpolate and there is no blurring filter in place that may reduce the resolution.
Note that we needed to enlarge the picture to 400% to see the differences.

There is another aspect to look at for a comparison between the Leica Monochrom and the Leica RGB image produced by the M8 and M. The Siemens test chart I use has lines that are radially fanning out and have fuzzy edges. The monochrom image has reduced fuzzyness compared to the RGB image and this higher edge sharpness may enhance the sharpness impression of the monochrom images. The fact that the 18 Mp monochrom image has better resolution than the 24 Mp RGB image can be solely attributed to the lower amount of aliasing. Basically this is the same effect as using an acutance developer that enhances edge contrast at the limit of resolution of the film emulsion. The facts are that a ISO 100 film with an acutance developer shows higher resolution and edge contrast than a ISO 25 film with a fine grain developer because the details are lost in the somewhat muddy grain structure of the slower speed film.
The pixel size of the Monochrom is 6.8 micron and the pixel size of the M is 6 micron or 145 pixels per mm versus166 pixels per mm. This is a theoretical improvement of 14.5%. The sensor area (Monochrom versus M) however has an increase of 33%. The improvement in resolution is lower than the improvement in area. This again resembles the classic comparison between a 24x36mm negative and the 60x60mm negative. As a side line note the excellent performance of the M8.2 which has the same pixel size as the Monochrom, but a smaller sensor area.
The effect of the reduced amount of aliasing can then be estimated as a 20% increase in resolution. If this reasoning can be applied to the new M Monochrom we can expect an improvement in effective resolution by the same amount, but note that the law of diminishing returns can be applied here too.
The fact that the Monochrom image is composed of the same three layers as the RGB image of the M can be verified in any RAW developer program. One sees simply the three channels (or color planes) but with identical brightness values, producing a neutral grey tone. The size of the file is also an indication that the Monochrom image has three layers (Monochrom image 36.4 MB and M image 48.5 MB). The three layers of the Monochrom seem to have identical brightness values in the three color planes that the demosaicing algorithm produces. Every pixel on the sensor area that is overlaid with the standard Bayer CFA pattern. Every pixel has only one parameter and that is the brightness value of the object point that is captured by the sensor. The red, blue and green colors have to be added later in the demosaicing process by the specific location of every pixel or at least the starting pixel in the uppermost left corner of the sensor. There are many different ways to reconstruct the color from a Bayer array and in the more sophisticated algorithms edge effects and fringing (sudden jumps in hue) are also considered in the final interpretation.
It seems that Leica uses the same processing inside the camera for the Monochrom and the M because the output file is still composed of the information per pixel in the full sensor matrix. Remember that the sensor area is a flat array of photo-sensitive pixels and that the Bayer CFA produces the RGGB pattern of pixels that reside on the same plane. For the sensor it makes no difference whether there is a Bayer layer in place or not. The final output is an matrix of brightness values. I do not know if Leica uses some processing inside the camera before the file is generated. The file size is an indication that Leica uses the full sensor area for its monochrom image file. And the RGB layer in the Raw programs is an indication that the program still assumes that the RGGB pattern of brightness values has to be reconstructed. Because there are color hues to reconstruct, it seems that the calculation of edge effects and reduction of fringing may enhance the resolution. But this is purely a guess.
Leica has recently made the shift in marketing to position their products in the artistic and emotional sphere and is therefore rather frugal with technical information. Technical comparative testing is no longer the rule and this makes it difficult to sift the wheat from the chaff (or to separate truth from myth).