Noise Reduction and Sharpening Quiz

Luminance noise is random variation in brightness from pixel to pixel, which looks like a fine gray grain, much like film grain. Colored speckles and blotches are a separate problem, called color noise. Raw editors usually offer a separate slider for each.

Color noise shows up as random red, green, and magenta speckles and blotches, mostly in dark areas and at high ISO. It comes from the sensor's color channels disagreeing from pixel to pixel. Brightness variation alone is luminance noise, a different thing.

Noise reduction works by smoothing small pixel-level variation, and fine detail like skin pores, hair, and fabric is also small pixel-level variation. Pushed too far, it erases that texture along with the noise, leaving surfaces that look waxy or plastic.

Every digital capture is slightly soft: the anti-aliasing filter blurs a little on purpose, and demosaicing interpolates color between pixels. Capture sharpening restores that lost crispness at the start of editing. It cannot rescue missed focus or motion blur, and it is not meant to create a stylized look.

A small web image viewed at a distance needs less sharpening than a large print seen close up. Output sharpening compensates for the softening introduced by the output medium and viewing conditions, so the correct amount depends entirely on those final variables.

Unsharp mask generates a blurred version of the image and adds the difference between the original and the blur back onto the original with a specified strength. The result amplifies local contrast at edges, making them appear crisper.

Deconvolution sharpening models the blur as a mathematical kernel and applies the inverse operation to reverse it. By working backward from the known blur characteristics, it can recover genuine detail that conventional edge-contrast sharpening simply amplifies.

A halo is a light or dark outline along an edge. Sharpening makes the light side of an edge lighter and the dark side darker; when the radius is too large or the amount too strong, that boost spreads into a visible band. Too little sharpening only looks soft, it does not create halos.

Frequency separation divides an image into two components: a low-frequency layer holding broad tonal transitions and gradients, and a high-frequency layer holding fine detail and texture. Adjustments to one layer affect neither the other, allowing skin tone correction without disturbing texture.

Texture and clarity sliders target intermediate spatial scales between broad gradients and fine edge detail. Raising them increases the contrast of features like skin pores and fabric weave without adding the halos that high-frequency sharpening can produce.

Sharpening amplifies contrast at existing edges but cannot restore information that was never recorded. Severe motion blur during exposure produces no edge information at all; sharpening applied to such areas increases artifacts without recovering detail.

Noise reduction comes first so sharpening does not amplify the noise into crunchy grain. Capture sharpening then restores softness from the sensor and demosaicing. Output sharpening comes last because it depends on the final size and medium, such as a web image or a print.

Downsampling averages several neighboring pixels into one, and random noise partly cancels out when averaged. A full resolution image shrunk for the web or a small print therefore shows less visible noise than it does at 100 percent.

Shadows hold the least signal, so they carry the most noise relative to detail. Brightening them a lot brightens that noise too, so heavy shadow lifting usually needs extra noise reduction, or a mask so the reduction applies only where shadows were lifted. Raising saturation there would make color noise worse.

Noise reduction and sharpening trade against each other: removing all noise also removes fine detail, and maximum sharpness amplifies noise. The right balance depends on the output, since a small web image hides noise that a large print reveals.