What is JPEG XS?
JPEG XS, standardized as ISO/IEC 21122, is a low-complexity, low-latency image and video codec. It supports visually lossless and mathematically lossless operation at moderate compression ratios.
How JPEG XS works
JPEG XS is built for links that used to carry uncompressed video, where images must be compressed and reconstructed with very little buffering and simple hardware. A wavelet transform splits each component into bands, the coefficients are grouped into precincts that each cover a region of the image, and precincts are combined into slices that span the full image width. That structure allows the latency added by encoding and decoding to be confined to a fraction of a frame, typically a few lines, and precise rate control lets a stream fit a fixed-capacity link. The cost is a modest compression ratio: typical visually lossless ratios are 2:1 to 10:1 in RFC 9134 and 2:1 to 18:1 in the third edition of the standard, published in 2024, so a stream typically still needs between about one eighteenth and one half of the uncompressed bitrate. Through the second edition every picture was coded on its own; the third edition adds an optional temporal differential coding mode in which a picture can depend on the one before it.
Key facts
- Whether every picture decodes on its own depends on the profile. A codestream that does not use temporal differential coding (TDC) is a stand-alone picture, which is all the first and second editions allowed. The third edition adds TDC profiles in which decoding can depend on a frame buffer left by the previous codestream, so confirm the profile before assuming frame-accurate switching, editing, or recovery after packet loss.
- JPEG XS trades compression for delay. The latency added by encoding and decoding can be confined to a fraction of a frame, typically a few lines, while typical visually lossless ratios are modest: 2:1 to 10:1 in RFC 9134, and 2:1 to 18:1, or higher depending on the image, in the 2024 third edition. That latency covers the codec only: the VSF TR-08 recommendation for ST 2110-22 systems sets a target of about one tenth of a frame for end-to-end transmission, still excluding network transit and buffering for path differences, so confirm device-to-device delay with vendors. Treat the ratios as typical values, not limits, and confirm the bitrate and picture quality each device delivers for your formats and content; the codec is designed for paths that previously carried uncompressed video.
- Being able to decode JPEG XS does not make two devices interoperate over IP. RFC 9134 defines two RTP packetization modes, codestream and slice, and the sender declares one in the required
packetmodeparameter. In SDP offer/answer negotiation, a receiver that does not support the offered parameters must reject the session. Interoperability recommendations narrow the choice further: VSF TR-08, for SMPTE ST 2110-22, mandates codestream mode and lists the profiles, levels, sublevels, and formats a device must handle for each configuration it claims, so check that both devices claim a common configuration. - A JPEG XS codestream holds sample values with no color interpretation. Color primaries, transfer characteristics, frame rate, and interlace mode travel in separate video support and color specification boxes or in the transport’s signaling, so a pipeline must carry that information alongside the codestream.
When JPEG XS matters
Professional production and video-over-IP links can use JPEG XS when predictable delay matters more than maximum compression. Both endpoints must support the standard and sustain its comparatively high bitrate.