White balance test
Demosaicing test
Highlight recovery test
Shadow recovery test
Noise reduction test
Sharpening test
Test images download page
Demosaicing test
Typical Bayer pattern sensors are able to detect only one color per pixel (red green or blue). Demosaicing algorithm interpolates the colors creating a full color image. A perfect demosaicing method should give great resolving power, sharp, still not jagged edges, small noise (see noise reduction test for comparison) and no artifacts (wrong colors, wrong interpolation directions etc.)
What we are looking at?
The default view is a crop from in camera jpeg. It shows the results from a decent DSLR (Konica Minolta dynax 7D). Every tested application shows a bit different contrast, colors and brightness - this is not the result of demosaicing algorithm but color rendering.White balance was always manually set to 9900K what corresponds to the color temperature of the LCD (yes - the pattern was displayed on LCD and then photographed). Images were not post processed (all noise reduction, sharpening and other tools were off) but it was upscaled 200% using ImageMagick.
Linux software
Bibble Pro 5.0.1DCRaw 8.88
Lightzone 3.6
Raw Therapee 2.4.1 allows to select from three demosaicing algorithms: HPHD, EAHD and 4 color VNG, and to set number of iterations of color smoothing for all of them. I present results of pure interpolation (HPHD(0), EAHD(0) and 4VNG(0)) and default number of color smoothing iterations set by designer (HPHD(1), EAHD(2) and 4VNG(2)).
Non-linux software
Aperture 2.1.2 I present only v2 of its demosaicing algorithm with default settingsCapture One Pro 5.1
DxO Optics Pro 6.0
Photoshop Lightroom 3 (beta 2) .
Photostudio Darkroom 1.5
SilkyPix Developer Studio Pro 4.1 allows to tweak the demosaicing sharpness. I present the default one.
Other
DCB - this demosaicing algorithm is the result of my hobby project (yes, I wrote it myself). DCB It is not implemented in any software. If you are interested, please visit this website. I am open for any kind of discussion.
![]() |
|||
| *** | *** | *** | *** |
| *** | *** | *** | *** |

