What demosaic algorithm is best for DNG to B&W conversion?

I figure that at some point someone has devised a demosaic algorithm that is optimal for converting a Bayer DNG to a B&W photograph. But what is that algorithm called, and what image processing software offers it?

dcraw has a "document mode" that has surprisingly good results but it doesn't merge pixels... there is no demosaicing.

I figure that at some point someone has devised a demosaic algorithm that is optimal for converting a Bayer DNG to a B&W photograph.

Well "optimal" is in the eye of the beholder.

Your idea of an optimal conversion may not be my idea of an optimal conversion. So on this notion the idea of an optimal conversion flounders.

On a fundamental level, demoisaicing algorithms are concerned with generating RGB images from RAW data.

And B&W image processing, even in the days of film, involved color filtering.

With film you actually did use a physical filter when you shot in order to get different contrast effects for different colors. In digital, normally, you'd work from a color image and apply a filter of choice (or any other processing) to get the effect you want.

The point is that B&W processing is almost entirely something you do now from a color base image (implicit even if you shoot RAW).

Even if you attempt to do this from RAW, you must construct an implicit full color image in order to apply a color filter-type of effect (something I'd regard as pretty much essential for serious B&W work). A color filter, by definition, requires a color source image.

The only alternative is rarely done these days with digital - shooting with a physical color filter.

But what is that algorithm called, and what image processing software offers it?

There is no single algorithm that would be universally considered optimal, but I can't say I've heard of an algorithm specifically designed to facilitate only B&W work. As I've explained, it's a self defeating idea to not work from a base color image these days.

• But if a conversion is first done to color (RGB) and then to B&W the emphasis will initially making the color look "good", which may adversely affect the luminosity. Surely there is a way to avoid the sometimes elaborate operations that companies use to adapt Bayer patterns to different types of scenes and whatnot. May 14, 2017 at 14:59
• No. The emphasis is not on making the color "look good", but to be as accurate as possible with respect to the color response of the sensor. "Looking good" is for secondary effects, like saturation, white balance, tone, sharpening, contrast, tone curve. RGB demosaicing is a necessity for generating good luminosity data. May 14, 2017 at 15:06
• Have you done much reading about how demosaicing works? There are several ways to do it. Some look really bad, some look better. It depends on the scene type. So I am correct, the demosaicing method is chosen to make the photo look good. And the choice usually takes color into account which does not apply in the case of a B&W photo. May 14, 2017 at 15:13
• Yes, I am quite familiar with demoisaicing algorithms. They are not chosen "to make the photo look good", they are generally chosen to minimize artifacts of interpolation, something that very careful expert evaluation of the algorithms decides. "Looking good" is a subjective quality of an image viewed as a complete composition, not at the pixel level of demosaicing. May 14, 2017 at 15:19
• @iSelfy "...But if a conversion is first done to color (RGB) and then to B&W the emphasis will initially making the color look "good", which may adversely affect the luminosity." The instant you capture the scene with a Bayer masked sensor the full luminosity has already been irretrievably lost to the light absorbed and reflected by the Bayer mask. May 14, 2017 at 22:26