unclipping

Unclipping is the process of restoring information that has been lost when a signal or image is recorded beyond the maximum representable value of a device, causing clipping or saturation. The aim is to approximate the original, undistorted signal or scene as closely as possible.

In audio, clipping occurs when the amplitude of a waveform exceeds a system’s headroom, producing distortion. Unclipping, or declipping, uses algorithms to estimate the removed portions of the waveform. Approaches include time-domain restoration that enforces signal constraints and uses priors, frequency-domain methods that exploit sparsity or spectral structure, and iterative reconstruction. The effectiveness depends on how much and what is clipped, and the level of prior knowledge about the original signal; artifacts can appear if the restoration is uncertain.

In images and video, clipping happens when pixel values reach the maximum or minimum of the dynamic range, causing loss of detail in highlights or shadows. Unclipping techniques include high dynamic range (HDR) imaging to merge multiple exposures, exposure or tone mapping to recover detail, and computational methods such as inpainting or deconvolution to infer plausible values for saturated regions. Recovery is often partial and relies on assumptions about texture, structure, or neighboring pixels; severe clipping may be unrecoverable.

In sensing and measurement, sensors can saturate or clip, leading to biased data. Unclipping in these contexts may involve calibration, modeling of the clipping behavior, or combining data from multiple channels or exposures to reconstruct the underlying signal.

See also: declipping, clipping distortion, high dynamic range imaging, inpainting.