Lossykodning

Lossykodning, also known as lossy encoding, is a data compression technique that reduces file size by selectively discarding information deemed unnecessary for the intended use. In contrast to lossless encoding, where the original data can be reconstructed exactly, lossykodning accepts some loss of fidelity in exchange for smaller representations. The discarded information is typically chosen according to a perceptual model that estimates what a human observer would not notice or would find difficult to distinguish.

Most lossy codecs use transform coding and quantization to separate signal into components, then quantize and encode the most perceptually important parts while discarding finer details. The pipeline often includes a transformation (such as a discrete cosine transform in images and audio, or a MDCT in audio/video), a quantizer, and entropy coding. The degree of loss is controlled by a quality or bitrate parameter, and rate-distortion tradeoffs determine the balance between file size and perceived quality.

Lossy encoding underpins most consumer media, including image formats like JPEG, audio formats such as MP3 and AAC, and video formats including MPEG-4, H.264, and HEVC. It is widely used for streaming, broadcasting, and storage where bandwidth or capacity is limited.

Disadvantages include the introduction of artifacts such as blockiness, ringing, or temporal smearing, and the fact that exact recovery of the original data is impossible. Artifacts and quality depend on the content, the codec, and the chosen bitrate. Quality assessment combines objective metrics such as PSNR and SSIM with subjective listening tests.

This approach contrasts with lossless compression, which preserves exact data at the cost of larger files. Lossy coding continues to evolve with advances in perceptual models and machine learning-based codecs that aim to improve quality at lower bitrates.