The VIPS file format
File format — The VIPS file format
VIPS 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
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 VIPS file format.
VIPS comes with a command-line program called vipsedit which is useful for destructively changing fields in a vips image. The vipsheader program can be used to extract any metadata.
VIPS 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 fields in the VIPS 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.
Table 2. The VIPS header
|0 -- 3||VIPS magic number: 08 f2 a6 b6, or b6 a6 f2 08|
|4 -- 7||int32||
||Width of image, in pixels|
|8 -- 11||int32||
||Height of image, in pixels|
|12 -- 15||int32||
||Number of image bands|
|16 -- 19||Unused|
|20 -- 23||VipsBandFormat||
|24 -- 27||VipsCoding||
|28 -- 31||VipsInterpretation||
|32 -- 35||float32||
||Horizontal resolution, in pixels per millimetre|
|36 -- 39||float32||
||Vertical resolution, in pixels per millimetre|
|40 -- 47||Unused|
|48 -- 51||int32||
||Horizontal offset of origin, in pixels|
|52 -- 55||int32||
||Vertical offset of origin, in pixels|
|56 -- 63||Unused|
The image data
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, VIPS will automatically byte-swap for you.
VIPS 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.
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.
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
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 VIPS image, and vipsedit --setext some_file.v < file.xml to replace the XML.