Nearlossless

Near-lossless refers to a class of data compression methods that permit a small, bounded amount of information loss, lying between lossless and lossy compression. In near-lossless schemes, the reconstruction of the original data is not exact, but the distortion is constrained by a user-defined bound, such that the maximum error in the reconstructed data does not exceed a specified threshold. The bound is often expressed in terms of peak signal-to-noise ratio (PSNR), mean squared error (MSE), or perceptual difference measures, and it may be defined per sample, per color channel, or per block.

In image and video compression, near-lossless modes are used to reduce data size while preserving visual fidelity. A common implementation is in JPEG-LS, an ISO/IEC 14495 standard that supports both lossless and near-lossless compression. In JPEG-LS, a near-lossless parameter quantifies the allowable difference between the original and decoded samples; NL = 0 yields lossless, NL > 0 yields increasing error bounds. Other codecs may offer similar bounded-distortion modes.

Near-lossless is especially relevant when exact reproducibility is either unnecessary or impractical, such as archival display, medical imaging where strict similarity bounds are required, or streaming where bandwidth is constrained but perceptual quality must be maintained. Critics note that the practical equivalence of near-lossless across different codecs depends on the chosen color space, quantization, and error metric.

There is no single universal definition or standard for near-lossless; it is a descriptive term applied to various schemes that guarantee a bounded reconstruction error rather than exact recovery.