Advanced Editing Tools

Noise Reduction

When the amount of light that reaches the sensor is reduced, the variation in response for each pixel increases. We see this as an increase in both luminance and color noise.  Even though full frame cameras have larger sensors and correspondingly larger pixels to gather more light than what is used on smartphones, reducing image noise is often a game changer when it comes to achieving ultimate image quality.  

Historically, noise reduction has been achieved by "smoothing" the signal  (i.e., numerically averaging) across adjacent pixels.  The consequence is a reduction in image sharpness.  With ever increasing computation power available in both desktop and laptop computers (now via dedicated graphics processors) more advanced AI or artificial intelligence base approaches have become available.  Several 3rd party plugin packages, including DXO's DeepPrime, ON1's NoNoise and Topaz's DeNoise have become available which reduce noise using more sophisticated algorithms.  

In April 2023, Lightroom introduced its own AI based denoising tools with its 12.3 release.  It is specifically developed to maximize image quality by working directly, and currently only, from raw files which have not been demosaiced.  There are several initial reviews which confirm excellent performance (See references below). What differentiates it from other noise reduction tools is the absence of introduced artifacts. In addition to noise reduction, color preservation/recovery has also been reported. 

I've seen impressive noise reduction and color capture performance using Lightroom's new AI based denoise applied to noisy, ISO 3200 micro four thirds camera data (12MP image) as a test case.  In this example the foliage is mixed, with some green and some withered and brown.  In addition to reducing noise, Lightroom AI denoise is able to correctly recover color for the green ivy.  This allowed for follow-on editing (increased contrast, decreased highlights, opened shadows, increased vibrance and sharpening) resulting in a very usable image.