Data compression that achieves high ratios by permanently discarding information judged perceptually or statistically less important, guided by rate-distortion theory and models of human vision and hearing; the basis of virtually all deployed image, audio and video coding — JPEG, MP3, AAC, Opus, H.264/HEVC/AV1 — where transform coding, quantisation and entropy coding together trade reconstruction fidelity against bitrate, in contrast to lossless methods that guarantee exact reconstruction.
Semantic Classification
Content
Definition
Lossy compression reduces data size by throwing information away — irreversibly. Where Lossless Compression exploits only statistical redundancy and reconstructs the original bit-for-bit, lossy methods additionally exploit irrelevance: components of a signal that a human observer will not miss. The theoretical foundation is Shannon’s rate-distortion theory, which characterises the minimum bitrate achievable for a given tolerated distortion; practical codecs approach this bound using perceptual models — the human eye’s lower sensitivity to fine colour detail than to luminance, the ear’s masking of quiet tones near loud ones — so that the discarded information is the least noticeable.
The canonical pipeline has three stages. A transform (the discrete cosine transform in JPEG and most video codecs, modified DCT in audio, wavelets in JPEG 2000) concentrates signal energy into few coefficients. Quantisation — the only lossy step — coarsens those coefficients, discarding precision according to perceptual weighting and a quality setting. Entropy coding then packs the quantised symbols losslessly. Video codecs add temporal prediction: motion estimation finds how blocks move between frames so that only residuals need coding, which is why H.264, HEVC and AV1 achieve ratios of 100:1 and beyond. The fidelity-bitrate trade-off is continuous, which is what makes Adaptive Bitrate Streaming possible: the same content is encoded at several quality rungs and clients switch between them as bandwidth fluctuates.
Lossy compression is appropriate for perceptual media consumed by humans; it is categorically wrong for executables, text, medical imagery used diagnostically, or any archival master, where lossless methods or no compression are mandated. Repeated lossy re-encoding compounds error — generation loss — which is why production workflows keep lossless or lightly compressed mezzanine files.
Current Landscape
Royalty questions and streaming economics continue to drive codec evolution: AV1 (AOMedia, royalty-free) and HEVC/VVC (licensed) compete for video, AVIF and JPEG XL for images, while Opus dominates real-time audio and AAC persists in broadcast. Recent milestones sharpen the picture:
-
AV2 released: the Alliance for Open Media finalised its royalty-free AV2 specification in late May 2026 (announced 9 June 2026), with prototype results showing roughly 28-33% bitrate reduction over AV1 at equivalent quality; an AOMedia member survey reports 53% plan adoption within a year and 88% within two.
-
H.267 kick-off: on 14 July 2025 JVET published requirements for the next-generation ITU-T/ISO video standard beyond VVC, targeting at least 40% bitrate reduction over VVC Main 10 for 4K content, with standardisation expected around 2028-2029; the Enhanced Compression Model (ECM) test model already demonstrates ~25% savings over VVC.
-
Learned codecs go mainstream in research: by 2025-2026 the best end-to-end neural video codecs (DCVC-RT class) match or beat VVC while running in real time on a single GPU, and JPEG AI / MPEG-AI standardisation tracks are formalising learned image and video coding — though these ship as model weights rather than fixed bitstream specifications.
-
Licensing pressure persists: Access Advance launched its Video Distribution Patent pool (covering HEVC, VVC, AV1 and VP9 content distribution) on 16 January 2025, with rates published July 2025 — keeping royalty economics central to codec selection for streaming at scale.
-
The same rate-distortion machinery now also underpins compression of neural network weights themselves for edge deployment.
Sources:
-
http://aomedia.org/press%20releases/Alliance-for-Open-Media-Releases-AV2-Codec/