The Noise Problem in Digital Photography
Digital noise, random variation in pixel brightness and colour, is an inherent characteristic of image sensors. It becomes most visible at high ISO settings, in shadow regions, and in images captured in low light. Noise reduction suppresses this random variation while preserving genuine image detail.
Traditional Noise Reduction Methods
Gaussian Blur
The simplest approach applies a Gaussian blur, averaging pixel values with their neighbours. This reduces random variation but also softens edges and fine detail.
Non-Local Means (NLM)
NLM finds similar patches across the entire image and averages those together. This preserves texture better than Gaussian blur but is computationally expensive.
Wavelet-Based Denoising
Wavelet transforms decompose an image into different frequency bands. Noise concentrates in high-frequency bands while detail spans multiple frequencies. Wavelet denoising thresholds the noise-dominated components while preserving structure. RawTherapee uses wavelet-based denoising effectively.
AI Noise Reduction: A Significant Advance
Modern AI noise reduction uses convolutional neural networks trained on pairs of clean and noisy versions of the same images. The network learns to predict what the clean version should look like, not by averaging pixels, but by understanding image structure semantically. Tools like Lightroom's Denoise and Topaz DeNoise AI can produce clean images from frames at ISO 12800 or higher.
How imagic Uses Quality Scoring for Noisy Images
imagic scores every image for noise as part of its AI Analyse step. High-noise images receive lower scores, helping identify frames that need intensive noise reduction in post. During burst grouping, the noise score contributes to selecting the frame with the cleanest sensor data.
Install imagic with pip install imagic to see noise scoring in action. The noise score visible in the Review step helps prioritise which images need dedicated AI noise reduction before final delivery.
Choosing the Right Tool
- darktable: Built-in denoising with multiple algorithms. Free, good quality.
- RawTherapee: Wavelet-based denoising with fine control. Excellent for preserving texture.
- Lightroom Denoise: AI-based, excellent quality, requires subscription.
- Topaz DeNoise AI: Best-in-class AI denoising, one-time purchase.