Mathematical basis function for image compression and analysis — foundation of DWT and modern VFX denoising. Supersedes Fourier in practice.
Wavelets have become the standard for image decomposition and reconstruction in modern VFX pipelines. Unlike the classic Fourier Transform, which decomposes an image into global frequencies, a wavelet works locally—it analyzes frequency and spatial position simultaneously. That's the key difference: you don't just see that there's noise in the image, but exactly where.
In practice, this means a huge advantage for denoising and image processing. Instead of filtering an entire image and destroying details in the process, you decompose the image using the Discrete Wavelet Transform (DWT) into multiple levels—coarse structures on one scale, fine details on another. Noise typically resides in the high-frequency components. You can selectively attenuate these while preserving large shapes and edges. This is more precise and produces fewer artifacts than classic Gaussian blur approaches. If you need to clean up an overexposed plate or a grainy tracking element, wavelet denoising works much more intelligently than brute-force blur filters.
The mathematical elegance lies in the fact that wavelets are adaptive basis functions—there are Daubechies wavelets, Morlet, Symlets, and many more. Each family has different properties: some preserve edges more sharply, others smooth more gently. In compositing, you'll quickly experiment here: for grain removal, you often need a different wavelet family than for image stabilization or for preparing tracking footage. Modern software like Nuke integrates wavelet-based tools directly (e.g., the Denoise node)—this is not by chance, but an acknowledgment that wavelets are simply the right mathematical language for digital image processing.
A practical tip: wavelet decomposition is computationally intensive but scalable. You can control the decomposition depth—more levels mean more analytical fineness, but also more render time. For 4K material, you're often well-served with 3–4 decomposition levels. And importantly: wavelet denoising requires parameters like Threshold and Decay—applying it blindly leads to plasticky-looking images. The art is to remove enough noise so it's not visibly disturbing, without sacrificing the natural film texture.