QOI (image format)
 | |
| Filename extension |
.qoi |
|---|---|
| Magic number | qoif (4 bytes, ASCII) |
| Developed by | Dominic Szablewski |
| Initial release | 24 November 2021 |
| Latest release | 1.0 5 January 2022 |
| Type of format | Lossless bitmap image format |
| Standard | Specification |
| Open format? | Yes |
| Free format? | Yes |
| Website | qoiformat.org |
The Quite OK Image Format (QOI) is a specification for lossless image compression of 24-bit (8 bits per color RGB) or 32-bit (8 bits per color with 8-bit alpha channel RGBA) color raster (bitmapped) images, invented by Dominic Szablewski and first announced November 24th 2021.[1]
Description
The intended purpose was to create an open source, faster, lossless compression method than PNG, that was easier to implement. A third party YouTube video explains how PNG and QOI compression are performed.[2] Figures specified in the announcement and the video claim 20–50x faster encoding, and 3–4x faster decoding speed compared to PNG, with compressed sizes similar to those of PNG.[1] The author has donated the specification to the public domain (CC0).
Software and language support
QOI is supported natively by ImageMagick,[3] IrfanView (as of version 4.60),[4] and FFmpeg (as of version 5.1).[5]
Community made plugins are available in GIMP, Paint.NET and XnView MP.[6]
There are also implementations for various languages such as Rust, Python, Java, C++, C# and more.[7] A full list can be found on the project's Git(Hub) repository README.
File format
Header
A QOI file consists of a 14-byte header, followed by any number of data “chunks” and an 8-byte end marker.
qoi_header {
char magic[4]; // magic bytes "qoif"
uint32_t width; // image width in pixels (BE)
uint32_t height; // image height in pixels (BE)
uint8_t channels; // 3 = RGB, 4 = RGBA
uint8_t colorspace; // 0 = sRGB with linear alpha
// 1 = all channels linear
};
The colorspace and channel fields are purely informative. They do not change the way data chunks are encoded.
Encoding
Images are encoded row by row, left to right, top to bottom. The
decoder and encoder start with {r: 0, g: 0, b: 0, a: 255} as the previous pixel value. An image is complete when all pixels specified by width * height have been covered. Pixels are encoded as:
- Run-length encoding of the previous pixel (
QOI_OP_RUN) - an index into the array of previously seen pixels (
QOI_OP_INDEX) - a difference compared to the previous pixel value in r,g,b (
QOI_OP_DIFForQOI_OP_LUMA) - Full r,g,b or r,g,b,a values (
QOI_OP_RGBorQOI_OP_RGBA)
The color channels are assumed to not be premultiplied with the alpha channel (“un-premultiplied alpha”). A running array[64] (zero-initialized) of previously seen pixel
values is maintained by the encoder and decoder. Each pixel that is seen by the encoder and decoder is put into this array at the position formed by a hash function of the color value.
In the encoder, if the pixel value at the index matches the current pixel, this index position is written to the stream as QOI_OP_INDEX. The hash function for the index is:
index_position = (r * 3 + g * 5 + b * 7 + a * 11) % 64
Each chunk starts with a 2- or 8-bit tag, followed by a number of data bits. The bit length of chunks is divisible by 8 - i.e. all chunks are byte aligned. All values encoded in these data bits have the most significant bit on the left. The 8-bit tags have precedence over the 2-bit tags. A decoder must check for the presence of an 8-bit tag first. The byte stream's end is marked with 7 0x00 bytes followed by a single 0x01 byte.
The possible chunks are:
┌─ QOI_OP_RGB ────┬─────────┬─────────┬─────────┐
│ Byte[0] │ Byte[1] │ Byte[2] │ Byte[3] │
│ 7 6 5 4 3 2 1 0 │ 7 .. 0 │ 7 .. 0 │ 7 .. 0 │
│─────────────────┼─────────┼─────────┼─────────│
│ 1 1 1 1 1 1 1 0 │ red │ green │ blue │
└─────────────────┴─────────┴─────────┴─────────┘
8-bit tag b11111110 (254)
8-bit red channel value
8-bit green channel value
8-bit blue channel value
The alpha value remains unchanged from the previous pixel.
┌─ QOI_OP_RGBA ───┬─────────┬─────────┬─────────┬─────────┐
│ Byte[0] │ Byte[1] │ Byte[2] │ Byte[3] │ Byte[4] │
│ 7 6 5 4 3 2 1 0 │ 7 .. 0 │ 7 .. 0 │ 7 .. 0 │ 7 .. 0 │
│─────────────────┼─────────┼─────────┼─────────┼─────────│
│ 1 1 1 1 1 1 1 1 │ red │ green │ blue │ alpha │
└─────────────────┴─────────┴─────────┴─────────┴─────────┘
8-bit tag b11111111 (255)
8-bit red channel value
8-bit green channel value
8-bit blue channel value
8-bit alpha channel value
┌─ QOI_OP_INDEX ──┐
│ Byte[0] │
│ 7 6 5 4 3 2 1 0 │
│────┼────────────│
│0 0 │ index │ (Range: 0..63)
└────┴────────────┘
2-bit tag b00
6-bit index into the color index array: 0..63
A valid encoder must not issue 2 or more consecutive QOI_OP_INDEX
chunks to the same index. QOI_OP_RUN should be used instead.
┌─ QOI_OP_DIFF ───┐
│ Byte[0] │
│ 7 6 5 4 3 2 1 0 │
│────┼───┼───┼────│
│ 0 1│dr │dg │ db │ (Range: 64 .. 127)
└────┴───┴───┴────┘
2-bit tag b01
2-bit red channel difference from the previous pixel -2..1
2-bit green channel difference from the previous pixel -2..1
2-bit blue channel difference from the previous pixel -2..1
The difference to the current channel values are using a wraparound operation, so 1 - 2 will result in 255, while 255 + 1 will result in 0.
Values are stored as unsigned integers with a bias of 2. E.g. −2 is stored as 0 (b00). 1 is stored as 3 (b11). The alpha value remains unchanged from the previous pixel.
┌─ QOI_OP_LUMA ───┬─────────────────┐
│ Byte[0] │ Byte[1] │
│ 7 6 5 4 3 2 1 0 │ 7 6 5 4 3 2 1 0 │
│────┼────────────┼────────┼────────│
│1 0 │ diff green │dr - dg │db - dg │ (Byte[0] Range: 128 .. 191)
└────┴────────────┴────────┴────────┘
2-bit tag b10
6-bit green channel difference from the previous pixel -32..31
4-bit red channel difference minus green channel difference -8..7
4-bit blue channel difference minus green channel difference -8..7
The green channel is used to indicate the general direction of change and is encoded in 6 bits. The red and blue channels (dr and db) base their diffs off of the green channel difference. I.e.:
dr_dg = (cur_px.r - prev_px.r) - (cur_px.g - prev_px.g)
db_dg = (cur_px.b - prev_px.b) - (cur_px.g - prev_px.g)
The difference to the current channel values are using a wraparound operation, so 10 - 13 will result in 253, while 250 + 7 will result in 1.
Values are stored as unsigned integers with a bias of 32 for the green channel and a bias of 8 for the red and blue channel. The alpha value remains unchanged from the previous pixel.
┌─ QOI_OP_RUN ────┐
│ Byte[0] │
│ 7 6 5 4 3 2 1 0 │
│────┼────────────│
│1 1 │ run │ (Range: 192 .. 253)
└────┴────────────┘
2-bit tag b11
6-bit run-length repeating the previous pixel: 1..62 111101 32 16 8 4 1
The run-length is stored with a bias of −1. Note that the runlengths 63 and 64 (b111110 and b111111) are illegal as they are occupied by the QOI_OP_RGB and QOI_OP_RGBA tags.[8]
References
- ^ a b "Lossless Image Compression in O(n) Time". Phoboslab.org. 2021-10-24. Retrieved May 1, 2022.
- ^ Reducible (March 3, 2022). "How PNG Works: Compromising Speed for Quality". YouTube.com. Retrieved May 1, 2022..
- ^ "ImageMagick - Image Formats". Retrieved May 4, 2022.
- ^ "History of IrfanView Changes/Versions". www.irfanview.com. Retrieved 2022-05-10.
- ^ "FFmpeg Changelog - Gitweb". ffmpeg.org. Retrieved 2022-07-13.
- ^ James Hein. "Moving images to the next level". Bangkok Post. Retrieved April 1, 2022.
- ^ Simon Sharwood. "Developer creates 'Quite OK Image Format' – but it performs better than just OK". The Register.
- ^ Szablewski, Dominic (2022-01-05). "The Quite OK Image Format Specification" (PDF). Retrieved 2022-06-05.
This article incorporates text from this source, which is in the public domain.
External links
- Format website: C Source code and benchmark results
- 1 page PDF specification
- GitHub repository (including C implementation)
