PhotoLab offers four noise reduction algorithms.
The function is called “DxO Denoising Technologies” (sic).

* High Quality, conventional noise processing to be manually adjusted when post-processing raster images (usually JPEG)
* HQ (fast) is a very powerful processing to be applied by default on raw photos.
PhotoLab’s preview window (the main window that shows corrections) includes HQ processing all the time
* PRIME (Probabilistic Raw IMage Enhancement) is known to preserve details and colors. It restores depth to unblocked shadow areas. Its application is stunning on photos taken with older cameras.
* DeepPRIME is more subtle than PRIME. This improved algorithm better preserves color rendering, whether dull or saturated, and enhances micro details

PRIME / DeepPRIME can only be applied on raw photos.
They are available with PhotoLab Elite Edition.

Click to enlarge – Alt+ < to return

DeepPRIME requires a muscular configuration and a compatible graphics card. The export times relative to PRIME are divided by 1.5 to 3.5. On the other hand, they risk being very long if the GPU is not invoked.
Transferring the DeepPRIME demosaicing load to the GPU frees up the CPU to continue working on a stable machine. This choice is made by DxO at the first launch of PhotoLab.

The Edit / Preferences – Performances – Selection between CPU or Graphics Card menu is used to neutralize or force the activation (it’s worth trying) of incompatible graphics cards. Otherwise leave it on Auto.

A tip for users with a low power machine.
– Export overnight with the option “Export in DNG with optical corrections and DeepPrime”
– Develop these photos by benefiting from the noise processing carried out (permanent display)

It is recommended to leave HQ (fast) noise reduction active at all times. The software divides the image into areas according to their density. It will process them as closely as possible depending on the characteristics of the camera, ISO sensitivity and luminance, or even apply no noise reduction at all.

It may be a good idea to systematically process with DeepPRIME all high ISO photos as well as those for which shadows have been planned.
– For scenes with high light dynamics, DeepPRIME cleans up shadows while preserving colors
– (very) dark scenes are by nature unfavorable to colors. DeepPRIME preserves them but does not work miracles if the colors are too narrow

PRIME / DeepPRIME processing consumes a lot of resources, it is not possible to display its effect permanently.
Its action is previewed in a thumbnail whose selection frame can be moved.

Click to enlarge – Alt+ < to return

When exporting the image, DeepPRIME noise reduction is performed at the very beginning of the development process, even before demosaicing. This is done under better conditions and the result is immediately more precise.

Unlike competing software whose noise reduction algorithms degrade the image, Optical Sharpness and MicroContrast corrections are performed on a better basis.

On particularly dirty images the output file size of the DeepPRIME version is smaller.
As an indication, This is the size of the different files of the same JPEG image.

6.4 Mb2.7 Mb2.2 Mb1.6 Mb

DeepPRIME can highlight skin defects (points, freckles, etc.). Use a negative microContrast Local Adjustment mask.

It is possible to filter the thumbnails of images processed with DeepPRIME, PRIME, and those with HQ reduction alone.

Click to enlarge – Alt+ < to return

About the first release of the PRIME algorithm:

The recommended power (by the magic wand) of the Luminance blurs too much detail. My own preset setting, “0- Standard”, is set to 30 which still effectively attenuates the fine grain.  

Note that the luminance value 0 still removes noise.

PRIME’s Labyrinth setting (Maze artifact) change the Labyrinth pattern visibility by counteracting the phenomenon which may appear on a frame with a frequency different from that of the DC sensor pixels.Previous versions of DxO Optics Pro processed it automatically, but at the price of general reduced acutance details to fight against a marginal phenomenon.
< See here > – All rights reserved – tous droits réservés.

«             »