The JP2 file (.jp2, .jpf, .j2k)

JP2 is the file format of JPEG 2000, an image standard the Joint Photographic Experts Group published in 2000 as a more capable successor to their original JPEG. Instead of cutting the picture into 8 × 8 blocks, it uses a wavelet transform that describes the whole image at several resolutions at once. The same standard covers lossless and lossy compression, very high bit depths, huge images and many colour channels, and lets a reader pull a smaller or lower-quality version out of a file without decoding all of it.

It never replaced JPEG for everyday photos or the web. Encoding and decoding cost more processing power, cameras and browsers stayed with JPEG, and for years the patent position was unclear, although the committee's aim was a royalty-free core. Instead, JPEG 2000 found homes where its features matter more than compatibility:

• Digital cinema. The Digital Cinema Initiatives specification chose JPEG 2000 in 2004, and the films shipped to cinemas as Digital Cinema Packages (DCPs) store every frame as a JPEG 2000 image.
• Libraries and archives. The US Library of Congress holds over 200 million JP2 files, including the scanned pages of its newspaper digitisation programme, and many archives keep lossless JP2 masters.
• Medical imaging. The DICOM standard lets scanners and image archives store images as JPEG 2000, lossless or lossy.
• PDFs and science. PDF can embed JPEG 2000 images, and space missions such as NASA's HiRISE camera on the Mars Reconnaissance Orbiter publish their images as JP2.

1. 1Colour transform. RGB is usually converted to brightness and two colour-difference channels: with an exact integer transform for lossless files, or the usual YCbCr-style transform for lossy ones.
2. 2Tiles. The image can be cut into rectangular tiles that are coded separately. Small images, like the example above, are a single tile.
3. 3Wavelet transform. Each tile is split into a half-size version of itself plus three sets of detail (horizontal, vertical, diagonal). The half-size version is split again, and so on for several levels. The result already contains the picture at 1/2, 1/4, 1/8 of its size: that's where JPEG 2000's resolution scalability comes from.
4. 4Quantisation (lossy only). The wavelet coefficients are divided and rounded. Lossless files skip this step and use the integer 5/3 wavelet, so every pixel comes back exactly.
5. 5Block coding. The coefficients are cut into code-blocks (often 64 × 64) and coded bit plane by bit plane, most significant first, with an arithmetic coder. This method is called EBCOT.
6. 6Packets and layers. The coded data is arranged into packets by resolution, component, position and quality layer. Cutting the stream short gives a lower-quality or smaller image instead of a broken one.

The Library of Congress sums it up as "compress once, decompress many ways": from one master file, a server can send a thumbnail, a screen-sized view or one region at full resolution, decoding only the parts it needs.

JPEG 2000 is a multi-part standard. The parts you're most likely to meet as files:

JPEG 2000 remains a sound choice for a lossless master that also needs to be served at many sizes, as in archives and image servers. For anything people will open themselves, convert to JPG, PNG, WebP or AVIF.

From Safari 5 onwards, Safari was the only major browser that displayed JPEG 2000, and some image CDNs sent JP2 versions of pictures to Safari users only. Mozilla declined to add it to Firefox in 2009, and Chrome never supported it. Apple's release notes for Safari 18, in September 2024, list one line under Images: "Removed support for JPEG2000." Since then no current browser shows a JP2 file in a page or when it's opened directly.

• On a website: convert to JPEG, WebP or AVIF, or let your CDN choose a format; check it no longer sends JP2 to Safari.
• For zoomable archive images: keep the JP2 masters on the server and let an image server convert on the fly. Servers following the IIIF Image API, such as IIPImage, read JP2 and send JPEG to the browser:

<!-- a lossless JP2 master on the server, a JPEG in the page -->
<img src="https://images.example.org/iiif/3/map-1890.jp2/full/1200,/0/default.jpg"
     alt="Town map, 1890">

• On a Mac: double-click to open in Preview, then File > Export to save as JPEG or PNG. Apple's image framework still reads JPEG 2000 even though Safari no longer does.
• On Windows: there's no built-in viewer for JPEG 2000. GIMP, IrfanView (with its plug-ins) and XnView MP open JP2 files and can save them as JPG or PNG.
• On the command line: OpenJPEG, the open-source reference implementation, includes opj_decompress -i image.jp2 -o image.png and opj_compress -i image.png -o image.jp2 (lossless unless you set a ratio, for example -r 20). With ImageMagick: magick image.jp2 image.png.
• In Python: Pillow reads and writes JPEG 2000 through OpenJPEG: Image.open("image.jp2").save("image.png").
• Inspect one: opj_dump -i image.jp2 prints the boxes and codestream settings: size, tiles, wavelet levels, layers and progression order.

• The image is broken or missing in Safari. It used to work: Safari 18 removed JPEG 2000. Serve JPEG, WebP or AVIF instead.
• An app says the file is invalid. It may be a bare codestream (FF 4F at the start) renamed to .jp2, or a JPX file using Part 2 features. Check the first bytes, and try OpenJPEG, which reads both.
• It's slow to open. Large JPEG 2000 images take far more processing to decode than JPEGs of the same size. Tiled files with several quality layers let a viewer decode only what's on screen; one huge tile doesn't.
• Colours look wrong. The file may use an ICC profile, a colour space only JPX supports, or more than 8 bits per channel that a simple viewer squeezes badly. Convert with a tool that handles colour management.
• A PDF page shows a blank image. The PDF embeds JPEG 2000 through its JPXDecode filter (part of PDF since version 1.5) and the viewer lacks a decoder. Try another PDF reader.
• Security. JPEG 2000 decoders, OpenJPEG included, have had memory-safety bugs fixed over the years. Keep them updated, and check an uploaded file's declared dimensions before decoding it.