De-Noise 2: Statistical and Psychovisual Noise Grain Equalization

A 3-panel before and after comparison of Denoise 2 results.
Left: pre-denoise image, middle: statistically optimal noise grain equalization, right: pyschovisually optimized noise grain equalization.

New in StarTools 1.6, the Denoise 2 module offers an alternative aesthetic to the venerable classic Denoise module, yet with all of its local signal quality Tracking power intact.

Denoise 2 is an acknowledgement of the "two schools" of noise reduction prevalent in astrophotography; there are those who like smooth images with little to no noise grain visible, and there are those who find some noise grain acceptable (or even desirable!) for the purpose of creating visual interest and general aesthetics (much like noise grain is added for a "filmic" look in CGI). The classic Denoise module caters to the former, while the Denoise 2 module caters to latter.


Usage

Denoise 2 is centered around a single, intuitive concept; noise grain equalization. That is, with a single intuitive parameter, an "acceptable" noise grain visibility threshold is specified above which noise grain is visually acceptable, and below which noise grain is visually unacceptable. The result is a final image that appears to have a constant level of noise grain, no matter how it was processed or stretched from the time it was linear.

The equalization of noise grain across the entire image is an important aspect of Denoise 2, as doing so will no longer draw the attention of a viewer to areas of low SNR (Signal-to-Noise Ratio); it makes the image appear as if it has a constant SNR across the entire image, precisely as most viewers are used to in other types of media (e.g. terrestrial daylight photography and cinematic sequences).

The level of "still acceptable" noise grain, is set by the 'Grain Removal' parameter. Higher values remove more grain to the point of blurring the image.

'Grain Limit Detail' and 'Grain Limit Color' set the largest visible noise grain size, for detail (luminance) and color respectively, that Denoise 2 should target. Beyond these limits, Denoise 2 will leave detail wholly intact. These parameters are derived from the 'Grain Size' parameter in the setup screen. The 'Grain Size' parameter in the setup screen merely helps establish a visual baseline (starting point); 'Grain Limit Detail' and 'Grain Limit Color' override this.

'Grain Limit Detail' and 'Grain Limit Color' set the largest visible noise grain size, for detail (luminance) and color respectively, that Denoise 2 should target.

Denoise 2 further - optionally - exploits the human visual system's poor ability to distinguish noise grain in the context of other detail, while also catering to its preference for acutance (a subjective perception of sharpness that is related to the edge contrast in an image). To this end, a psychovisual support image can be synthesised and evaluated. The Psychovisual support image should not be mistaken for a simple luminance mask blending the original and denoised image, rather it serves as an input to acutance modelling and signal-to-noise ratio estimation during denoising.

The 'Mode' parameter toggles between statistically correct grain removal and psychovisual grain removal. When 'Mode' is set to 'Statistical' the 'PV Detail', 'PV Support Gamma' and 'PV Support Area' are unavailable as these are for the psychovisual ("PV") grain removal mode only.

When the 'Mode' parameter is set to 'Psychovisual', the 'PV Detail', 'PV Support Gamma' and 'PV Support Area' become available. 'PV Detail' governs what detail sizes Denoise 2 should look at when deciding whether something is visually "busy". 'PV Support Gamma' governs a non-linear transformation of the support image, non-linearly stretching the 'busyness' map. Increasing this parameter increases the strength of the grain preservation in busy areas. Finally, 'PV Support Area' specifies how 'busyness' psychovisually affect neighboring pixels in an area. Increasing this value will make grain preservation 'bleed' into neighboring pixels, preserving more grain (and detail).

Psychovisual noise reduction and detail preservation is - by its very nature - an imprecise process. While evaluating the resulting Denoised image, it is important to try to keep a "bird's eye view" of the noise grain in your image; it is important to imagine someone looking at your denoised image for the first time with "fresh eyes" and no prior knowledge of the original noisiness of your image. "Pixel-peeping" (e.g. looking at the image at increased zoom levels than native 100% zoom) will reveal noise much more readily than normal, rather than letting the psychovisual tricks do their work at native resolution (in that sense the effect is similar to that of quantization error dithering).

As is the case with the classic Denoise module, the Denoise 2 module was primarily designed for targeting Poissonian ("shot") noise.

The Denoise 2 module relies entirely on the per-pixel statistics Tracking provides, and as such, it is not available for non-Tracked processing; the Denoise 2 cannot yield correct results if the dataset was not linear at the time of engaging Tracking.