The libvips file format

libvips has a simple, native file format. It’s very fast, there is no image size limit, and it supports arbitrary metadata. Although few other programs can read these images (though recent versions of ImageMagick do have basic support for the .v format), it can be useful as an intermediate format for command-line processing. For example:

$ vips invert input.tif t.v
$ vips gamma t.v output.tif

is faster than using .tif for the temporary intermediate image. This section documents the libvips file format.

libvips comes with a command-line program called vipsedit which is useful for destructively changing fields in a .v image. The vipsheader program can be used to extract any metadata.

libvips files come in three parts. First, there is a 64-byte header, containing an identifying magic number and a set of very basic fields, such as image width in pixels. Next, the image data is stored as a set of band-interleaved scanlines, from the top of the image to the bottom. Finally, after the pixel data comes an optional block of XML containing any extra metadata, such as an ICC profile or the EXIF data.

The header

The fields in the libvips header are always stored least-significant byte first (Intel ordering). Only the most basic information about the image is in the header: most metadata is stored in the XML extension block after the pixel data.

If the first four bytes of the file are in order 08 f2 a6 b6, the image data (see the next section) is stored in Intel byte order (LSB first) and will need to be swapped if read on a SPARC-style machine (MSB first). If the magic number is b6 a6 f2 08, the image data is in SPARC order and will need to swapped if read on an Intel-style machine. libvips does this swapping automatically.

The image data

If coding is set to VIPS_CODING_NONE, pixels are stored in native C format, that is, the native format of the machine that wrote the data. If you open a big-endian image on a little-endian machine, libvips will automatically byte-swap for you. libvips has 10 band formats, see VipsBandFormat. Image data is stored as a simple list of scanlines, from the top of the image to the bottom. Pixels are band-interleaved, so RGBRGBRGBRGB, for example. There is no padding at the end of scanlines.

If coding is set to VIPS_CODING_LABQ, each pixel is four bytes, with 10 bits for L* and 11 bits for each of a* and b*. These 32 bits are packed into 4 bytes, with the most significant 8 bits of each value in the first 3 bytes, and the left-over bits packed into the final byte as 2:3:3.

If coding is set to VIPS_CODING_RAD, each pixel is RGB or XYZ float, with 8 bits of mantissa and then 8 bits of exponent, shared between the three channels. This coding style is used by the Radiance family of programs (and the HDR format) commonly used for HDR imaging.

Other values of coding can set other coding styles. Use VIPS_IMAGE_SIZEOF_IMAGE() to calculate the size of the image data section.

The metadata

Following the image data is a chunk of XML holding a simple list of name-value pairs. Binary data is encoded with base64. Use vips_image_set() and friends to set and get image metadata.

You can use vipsheader -f getext some_file.v to get the XML from a libvips image, and vipsedit --setext some_file.v < file.xml to replace the XML.