- Notifications
You must be signed in to change notification settings - Fork93
Zig library for reading and writing different image formats
License
zigimg/zigimg
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
This is a work in progress library to create, process, read and write different image formats withZig programming language.
This library uses zig nominated2024.11.0-mach. To install usingzigup
:
zigup 0.14.0-dev.2577+271452d22
How to add to your project:
- Clone this repository or add as a submodule
- Add to your
build.zig
pub fn build(b: *std.Build) void { exe.root_module.addAnonymousModule("zigimg", .{ .root_source_file = b.path("zigimg.zig") });}
- Run this command in your project folder to add
zigimg
to yourbuild.zig.zon
zig fetch --save git+https://github.com/zigimg/zigimg.git
- Get the module in your
build.zig
file
constzigimg_dependency=b.dependency("zigimg", .{ .target=target, .optimize=optimize,});exe.root_module.addImport("zigimg",zigimg_dependency.module("zigimg"));
After you are done setting up, you can look at the user guide below.
To run the test suite, checkout thetest suite and run
- Checkout zigimg
- Go back one folder and checkout thetest suite
- Run the tests with
zig build
zig build test
Image Format | Read | Write |
---|---|---|
ANIM | ❌ | ❌ |
BMP | ✔️ (Partial) | ✔️ (Partial) |
Farbfeld | ✔️ | ✔️ |
GIF | ✔️ | ❌ |
ICO | ❌ | ❌ |
IFF | ✔️ | ❌ |
JPEG | ❌ | ❌ |
PAM | ✔️ | ✔️ |
PBM | ✔️ | ✔️ |
PCX | ✔️ | ✔️ |
PGM | ✔️ (Partial) | ✔️ (Partial) |
PNG | ✔️ | ✔️ (Partial) |
PPM | ✔️ (Partial) | ✔️ (Partial) |
QOI | ✔️ | ✔️ |
TGA | ✔️ | ✔️ |
TIFF | ❌ | ❌ |
XBM | ❌ | ❌ |
XPM | ❌ | ❌ |
- version 4 BMP
- version 5 BMP
- 24-bit RGB read & write
- 32-bit RGBA read & write
- Doesn't support any compression
- Support GIF87a and GIF89a
- Support animated GIF with Netscape application extension for looping information
- Supports interlaced
- Supports tiled and layered images used to achieve pseudo true color and more.
- The plain text extension is not supported
- Supports 1-8 bit, 24 bit, HAM6/8, EHB ILBM files
- Supports uncompressed, byterun 1 & 2 (Atari) compressed ILBM files
- Supports PBM (Deluxe Paint DOS) encoded files
- Supports ACBM (Amiga Basic) files
- Color cycle chunks are ignored
- Mask is not supported (skipped)
Currently, this only supports a subset of PAMs where:
- The tuple type is official (see
man 5 pam
) or easily inferred (and by extension, depth is 4 or less) - All the images in a sequence have the same dimensions and maxval (it is technically possible to support animations with different maxvals and tuple types as each
AnimationFrame
has its ownPixelStorage
, however, this is likely not expected by users of the library) - Grayscale,
- Grayscale with alpha
- Rgb555
- Rgb24 and Rgba32
- Bgr24 and Bgra32
- Rgb48 and Rgba64
- Everything is supported
- Support monochrome, 4 color, 16 color and 256 color indexed images
- Support 24-bit RGB images
- Support 8-bit and 16-bit grayscale images
- 16-bit ascii grayscale loading not tested
- Support all pixel formats supported by PNG (grayscale, grayscale+alpha, indexed, truecolor, truecolor with alpha) in 8-bit or 16-bit.
- Support the mininal chunks in order to decode the image.
- Not all images in Png Test Suite is covered but should be good enough for now.
- Support 24-bit RGB (8-bit per channel)
- Missing 48-bit RGB (16-bit per channel)
- Imported fromhttps://github.com/MasterQ32/zig-qoi with blessing of the author
- Supports uncompressed and compressed 8-bit grayscale, indexed with 16-bit and 24-bit colormap, truecolor with 16-bit(RGB555), 24-bit or 32-bit bit depth.
- Supports reading version 1 and version 2
- Supports writing version 2
- Indexed: 1bpp (bit per pixel), 2bpp, 4bpp, 8bpp, 16bpp
- Grayscale: 1bpp, 2bpp, 4bpp, 8bpp, 16bpp, 8bpp with alpha, 16bpp with alpha
- Truecolor: RGB332, RGB555, RGB565, RGB24 (8-bit per channel), RGBA32 (8-bit per channel), BGR555, BGR24 (8-bit per channel), BGRA32 (8-bit per channel), RGB48 (16-bit per channel), RGBA64 (16-bit per channel)
- float: 32-bit float RGBA, this is the neutral format.
zigimg offers color and image functionality. The library is designed around either using the convenientImage
(orImageUnmanaged
) struct that can read and write image formats no matter the format.
Or you can also use the image format directly in case you want to extract more data from the image format. So if you find thatImage
does not give you the information that you need from a PNG or other format, you can use the PNG format albeit with a more manual API thatImage
hide from you.
Image
bundle a memory allocator andImageUnmanaged
does not. Similar tostd.ArrayList()
andstd.ArrayListUnmanaged()
in Zig standard library. For all the examples we are going to useImage
but it is similar withImageUnmanaged
.
It is pretty straightforward to read an image using theImage
struct.
You can use either a file path
conststd=@import("std");constzigimg=@import("zigimg");pubfnmain()!void {vargpa=std.heap.GeneralPurposeAllocator(.{}){};defer_=gpa.deinit();constallocator=gpa.allocator();varimage=tryzigimg.Image.fromFilePath(allocator,"my_image.png");deferimage.deinit();// Do something with your image}
or astd.fs.File
directly
conststd=@import("std");constzigimg=@import("zigimg");pubfnmain()!void {vargpa=std.heap.GeneralPurposeAllocator(.{}){};defer_=gpa.deinit();constallocator=gpa.allocator();varfile=trystd.fs.cwd().openFile(file_path, .{});deferfile.close();varimage=tryzigimg.Image.fromFile(allocator,file);deferimage.deinit();// Do something with your image}
conststd=@import("std");constzigimg=@import("zigimg");constimage_data=@embedFile("test.bmp");pubfnmain()!void {vargpa=std.heap.GeneralPurposeAllocator(.{}){};defer_=gpa.deinit();constallocator=gpa.allocator();constimage=tryzigimg.Image.fromMemory(allocator,image_data[0..]);deferimage.deinit();// Do something with your image}
For a single image, they are two ways to get access to the pixel data.
Yu can access the pixel data directly usingImage.pixels
.pixels
is an union of all supported pixel formats.
For RGB pixel formats, just use the pixel format enum value and addresss the data directly.
pubfnexample()void {// [...]// Assuming you already have an image loadedconstfirst_pixel=image.pixels.rgb24[0];}
For grayscale formats, you need to use .value to get the grayscale value. It can also contain the alpha value if you use the grayscale with alpha.
pubfnexample()void {// [...]// Assuming you already have an image loadedconstfirst_pixel=image.pixels.grayscale8Alpha[0];constgrayscale=first_pixel.value;constalpha=grayscale.alpha;}
For indexed formats, you need to first access the union value then either the indices or the palette. The palette color are stored in theRgba32
pixel format.
pubfnexample()void {// [...]// Assuming you already have an image loadedconstfirst_color_palette=image.pixels.indexed8.palette[0];constfirst_pixel=image.pixels.indexed8.indices[0];}
If you want to know the current pixel format useImage.pixelFormat()
.
You can use the iterator to get each pixel as the universalColorf32
pixel format. (32-bit floating ploint RGBA)
pubfnexample()void {// [...]// Assuming you already have an image loadedconstcolor_it=image.iterator();while (color_it.next())|color| {// Do something with color }}
In the case of anImage
containing multiple frames, you can useImage.animation
to get access to the animation information. UseImage.animation.frames
to access each indivial frame. Each frame contain the pixel data and a frame duration in seconds (32-bit floating point).
Image.pixels
will always point to the first frame of an animation also.
pubfnexample()void {// [...]// Assuming you already have an image loadedconstloop_count=image.animation.loop_count;for (image.animation.frames)|frame| {constrgb24_data=frame.pixels.rgb24;constframe_duration=frame.duration; }}
Image
has helper functions to help you get the right data to upload your image to the GPU.
pubfnexample()void {// [...]// Assuming you already have an image loadedconstimage_data=image.rawBytes();constrow_pitch=image.rowByteSize();constimage_byte_size=image.imageByteSize();}
You can query the image format used by a file or a memory buffer.
You can use either a file path
conststd=@import("std");constzigimg=@import("zigimg");pubfnmain()!void {constimage_format=tryzigimg.Image.detectFormatFromFilePath(allocator,"my_image.png");// Will print pngstd.log.debug("Image format: {}", .{image_format});}
or astd.fs.File
directly
conststd=@import("std");constzigimg=@import("zigimg");pubfnmain()!void {varfile=trystd.fs.cwd().openFile("my_image.gif", .{});deferfile.close();constimage_format=tryzigimg.Image.detectFormatFromFile(allocator,file);// Will print gifstd.log.debug("Image format: {}", .{image_format});}
conststd=@import("std");constzigimg=@import("zigimg");constimage_data=@embedFile("test.bmp");pubfnmain()!void {constimage_format=tryzigimg.Image.detectFormatFromMemory(allocator,image_data[0..]);// Will print bmpstd.log.debug("Image format: {}", .{image_format});}
Each 3 functions to write an image take a union of encoder options for the target format. To know the actual options you'll need to consult the source code. The active tag of the union determine the target format, not the file extension.
pubfnexample()!void {// [...]// Assuming you already have an image loadedtryimage.writeToFilePath("my_new_image.png", .{ .png= .{} });// Or with encoder optionstryimage.writeToFilePath("my_new_image.png", .{ .png= .{ .interlaced=true } });}
pubfnexample()!void {// [...]// Assuming you already have an image loaded and the file already createdtryimage.writeToFile(file, .{ .bmp= .{} });}
Ensure that you have enough place in your buffer before callingwriteToMemory()
pubfnexample()!void {// [...]// Assuming you already have an image loaded and the buffer already allocatedtryimage.writeToMemory(buffer[0..], .{ .tga= .{} });}
UseImage.create()
and pass the width, height and the pixel format that you want.
conststd=@import("std");constzigimg=@import("zigimg");pubfnmain()!void {vargpa=std.heap.GeneralPurposeAllocator(.{}){};defer_=gpa.deinit();constallocator=gpa.allocator();varimage=tryzigimg.Image.create(allocator,1920,1080,.rgba32);deferimage.deinit();// Do something with your image}
If you are not dealing with a image format, you can import your pixel data usingImage.fromRawPixels()
. It will create a copy of the pixels data. If you want the image to take ownership or just pass the data along to write it to a image format, useImageUnmanaged.fromRawPixelsOwned()
.
UsingfromRawPixel()
:
conststd=@import("std");constzigimg=@import("zigimg");pubfnmain()!void {vargpa=std.heap.GeneralPurposeAllocator(.{}){};defer_=gpa.deinit();constallocator=gpa.allocator();constmy_raw_pixels=@embedData("raw_bgra32.bin");varimage=tryzigimg.Image.fromRawPixels(allocator,1920,1080,my_raw_pixels[0..],.bgra32);deferimage.deinit();// Do something with your image}
UsingfromRawPixelsOwned()
:
conststd=@import("std");constzigimg=@import("zigimg");pubfnmain()!void {vargpa=std.heap.GeneralPurposeAllocator(.{}){};defer_=gpa.deinit();constallocator=gpa.allocator();constmy_raw_pixels=@embedData("raw_bgra32.bin");varimage=tryzigimg.ImageUnmanaged.fromRawPixelsOwned(1920,1080,my_raw_pixels[0..],.bgra32);// Do something with your image}
In the case you want more direct access to the image format, all the image formats are accessible from thezigimg
module. However, you'll need to do a bit more manual steps in order to retrieve the pixel data.
conststd=@import("std");constzigimg=@import("zigimg");pubfnmain()!void {vargpa=std.heap.GeneralPurposeAllocator(.{}){};defer_=gpa.deinit();constallocator=gpa.allocator();constimage_data=@embedFile("windows_rgba_v5.bmp");varstream_source=std.io.StreamSource{ .const_buffer=std.io.fixedBufferStream(image_data) };varbmp=zigimg.formats.bmp.BMP{};constpixels=trybmp.read(allocator,&stream_source);deferpixels.deinit(allocator);std.log.info("BMP info header: {}", .{bmp.info_header});}
For the curious, the program above generate the following output:
info: BMP info header: src.formats.bmp.BitmapInfoHeader{ .v5 = src.formats.bmp.BitmapInfoHeaderV5{ .header_size = 124, .width = 240, .height = 160, .color_plane = 1, .bit_count = 32, .compression_method = src.formats.bmp.CompressionMethod.bitfields, .image_raw_size = 153600, .horizontal_resolution = 2835, .vertical_resolution = 2835, .palette_size = 0, .important_colors = 0, .red_mask = 16711680, .green_mask = 65280, .blue_mask = 255, .alpha_mask = 4278190080, .color_space = src.formats.bmp.BitmapColorSpace.srgb, .cie_end_points = src.formats.bmp.CieXyzTriple{ .red = src.formats.bmp.CieXyz{ ... }, .green = src.formats.bmp.CieXyz{ ... }, .blue = src.formats.bmp.CieXyz{ ... } }, .gamma_red = 0, .gamma_green = 0, .gamma_blue = 0, .intent = src.formats.bmp.BitmapIntent.graphics, .profile_data = 0, .profile_size = 0, .reserved = 0 } }
You can useImage.convert()
to convert between pixel formats. It will allocate the new pixel data and free the old one for you. It supports conversion from and to any pixel format. When converting down to indexed format, no dithering is done.
pubfnexample()!void {// [...]// Assuming you already have an image loadedtryimage.convert(.float32);}
If you prefer, you can usePixelFormatConverter
directly.
pubfnexample(allocator:std.mem.Allocator)!void {constindexed2_pixels=tryzigimg.color.PixelStorage.init(allocator,.indexed2,4);deferindexed2_pixels.deinit(allocator);// [...] Setup your indexed2 pixel dataconstbgr24_pixels=tryzigimg.PixelFormatConverter.convert(allocator,&indexed2_pixels,.bgr24);deferbgr24_pixels.deinit(allocator);}
If you prefer more granular control to create an indexed image, you can use theOctTreeQuantizer
directly.
pubfnexample(allocator:std.mem.Allocator)!void {constimage_data=@embedFile("windows_rgba_v5.bmp");varimage=tryzigimg.Image.fromMemory(allocator,image_data[0..]);deferimage.deinit();varquantizer=zigimg.OctTreeQuantizer.init(allocator);deferquantizer.deinit();varcolor_it=image.iterator();while (color_it.next())|pixel| {tryquantizer.addColor(pixel); }varpalette_storage: [256]zigimg.color.Rgba32=undefined;constpalette=quantizer.makePalette(255,palette_storage[0..]);constpalette_index=tryquantizer.getPaletteIndex(zigimg.color.Rgba32.initRgba(110,0,0,255));}
You can get a color from a HTML hex string. The alpha component is always last. It also supports the shorthand version.
pubfnexample()!void {constrgb24=tryzigimg.color.Rgb24.fromHtmlHex("#123499");constrgba32=tryzigimg.color.Rgba32.fromHtmlHex("FF000045");constred_rgb24=tryzigimg.color.Rgb24.fromHtmlHex("#F00");constblue_rgba32=tryzigimg.clor.Rgba32.fromHtmlHex("#00FA");}
You can access predefined colors for any pixel format usingColors()
.
conststd=@import("std");constzigimg=@import("zigimg");pubfnmain()!void {constred_float32=zigimg.Colors(zigimg.color.Colorf32).Red;constblue_rgb24=zigimg.Colors(zigimg.color.Rgb24).Blue;}
While zigimg does not support ICC profile yet (see #36) it does support a variety of color models and color spaces. All color space and color model are done in 32-bit floating point. So if you are not usingColorf32
/float32
as your pixel format, you'll need to convert to that format first.
The following device-dependent color model are supported:
- HSL (Hue, Saturation, Luminance)
- HSV (Hue, Saturation, Value) or also known as HSB (Hue, Saturation, Brightness)
- CMYK (Cyan-Magenta-Yellow-Black)
The following device-inpendent color spaces are supported, with or without alpha:
- CIE XYZ
- CIE Lab
- CIE LCh(ab), the cylindral representation of CIE Lab
- CIE Luv
- CIE LCh(uv), the cylindral representation of CIE Luv
- HSLuv, a HSL representation of CIE LCh(uv) which is a cylindrical representation of CIE Luv color space
- Oklab
- Oklch, the cylindrical representation of Oklab
All color space transformation are done assuming a linear version of the color. To convert between gamma-converted and linear, you need to use any RGB colorspace and then calltoGamma()
ortoLinear()
, in this example I'm using bothsRGB
andBT709
(aka Rec.709).
You can use either the accurate version or the fast version. For example the sRGB transfer function is linear below a threshold and an exponent curve above the threshold but the fast version will use the approximate exponent curve for the whole range.
pubfnexample(linear_color:zigimg.color.Colorf32) {constgamma_srgb=zigimg.color.sRGB.toGamma(linear_color);constgamma_bt709=zigimg.color.BT709.toGammaFast(linear_color);constlinear_srgb=zigimg.color.sRGB.toLinearFast(gamma_srgb);constlinear_bt709=zigimg.color.BT709.toLinear(gamma_bt609);}
To convert to a device independant color space, you need first to use a reference RGB color space. Usually the most common for computer purposes issRGB
. Then each RGB colorspace has functions to convert from and to various color spaces. They support both non-alpha and alpha of the color space.
To a color space:
pubfnexample(linear_color:zigimg.color.Colorf32)void {constxyz=zigimg.color.sRGB.toXYZ(linear_color);constlab_alpha=zigimg.color.sRGB.toLabAlpha(linear_color);constlch_ab=zigimg.color.sRGB.toLCHab(linear_color);constluv_alpha=zigimg.color.sRGB.toLuvAlpha(linear_color);constlch_uv=zigimg.color.sRGB.toLCHuv(linear_color);consthsluv=zigimg.color.sRGB.toHSLuv(linear_color);constoklab=zigimg.color.sRGB.toOklab(linear_color);constoklch=zigimg.color.sRGB.toOkLCh(linear_color);}
When converting from a color space to a RGB color space, you need to specify if you want the color to be clamped inside the RGB colorspace or not because the resulting color could be outside of the RGB color space.
pubfnexample(oklab:zigimg.color.Oklab) {constlinear_srgb_clamped=zigimg.color.sRGB.fromOklab(oklab,.clamp);constlinear_srgb=zigimg.color.sRGB.fromOklab(oklab,.none);}
Converting each pixel individually will be tedious if you want to use image processing on the CPU. Almost all color space conversion offeran slice in-place conversion or a slice copy conversion. The in-place will reuse the same memory but interpret the color data differently. When you are conversion from a color space to a RGB color space, you need to specify if you want clamping or not.
Those conversions are only available with the alpha version of each color space.
pubfnexampleInPlace(linear_srgb_image: []zigimg.color.Colorf32)void {constslice_lab_alpha=zigimg.color.sRGB.sliceToLabAlphaInPlace(linear_srgb_image);// Do your image manipulation in CIE L*a*b*// Convert back to linear sRGB_=zigimg.color.sRGB.sliceFromLabAlphaInPlace(slice_lab_alpha,.clamp);// or without clamping_=zigimg.color.sRGB.sliceFromLabAlphaInPlace(slice_lab_alpha,.none);}pubfnexampleCopy(allocator:std.mem.Allocator,linear_srgb_image: []constzigimg.color.Colorf32)![]zigimg.color.Colorf32 {constslice_oklab_alpha=tryzigimg.color.sRGB.sliceToOklabCopy(allocator,linear_srgb_image);// Do your image manipulatioon in Oklab// Convert back to linear sRGBreturntryzigimg.color.sRGB.sliceFromOklabCopy(allocator,slice_oklab_alpha,.clamp);// Or without clampingreturntryzigimg.color.sRGB.sliceFromOklabCopy(allocator,slice_oklab_alpha,.none);}
CIE Lab, CIE Luv and Oklab have cylindrical representation of their color space, each color has functions to convert from and to the cylindrical version.
pubfnexample()void {constlab=zigimg.color.CIELab{ .l=0.12, .a=-0.23, .b=0.56 };constluv_alpha=zigimg.color.CIELuvAlpha { .l=0.4, .u=0.5, .v=-0.2, .alpha=0.8 };constoklab=zigimg.color.Oklab{ .l=0.67, .a=0.1, .b=0.56 };constlch_ab=lab.toLCHab();constlch_uv_alpha=luv_alpha.toLCHuvAlpha();constoklch=oklab.toOkLCh();constresult_lab=lch_ab.toLab();constresult_luv_alpha=lch_uv_alpha.toLuvAlpha();constresult_oklab=oklch.toOklab();}
To convert a single color, use theconvertColor()
function on theRgbColorspace
struct:
pubfnexample(linear_color:zigimg.color.Colorf32)void {constpro_photo_color=zigimg.color.sRGB.convertColor(zigimg.color.ProPhotoRGB,linear_color);}
If you want to convert a whole slice of pixels, useconvertColors()
, it will apply the conversion in-place:
pubfnexample(linear_image: []zigimg.color.Colorf32)void {constadobe_image=zigimg.color.sRGB.convertColors(zigimg.color.AdobeRGB,linear_image);}
If the target RGB colorspace have a different white point, it will do thechromatic adapdation for you using the Bradford method.
Here the list of predefined RGB color spaces, all accessible fromzigimg.color
struct:
BT601_NTSC
BT601_PAL
BT709
sRGB
DCIP3.Display
DCIP3.Theater
DCIP3.ACES
BT2020
AdobeRGB
AdobeWideGamutRGB
ProPhotoRGB
All predefined white point are accessed withzigimg.color.WhitePoints
. All the standard illuminants are defined there.
You can create your own RGB color space usingzigimg.color.RgbColorspace.init()
. Each coordinate is in the 2D version of the CIE xyY color space.
If you don't care about linear and gamma conversion, just ignore those functions in the init struct.
fnmyColorSpaceToGamma(value:f32)f32 {returnstd.math.pow(f32,value,1.0/2.4);}fnmyColorSpaceToLinear(value:f32)f32 {returnstd.math.pow(f32,value,2.4);}pubfnexample()void {pubconstmy_color_space=zigimg.color.RgbColorspace.init(.{ .red= .{ .x=0.6400, .y=0.3300 }, .green= .{ .x=0.3000, .y=0.6000 }, .blue= .{ .x=0.1500, .y=0.0600 }, .white=zigimg.color.WhitePoints.D50, .to_gamma=myColorSpaceToGamma, .to_gamma_fast=myColorSpaceToGamma, .to_linear=myColorSpaceToLinear, .to_linear_fast=myColorSpaceToLinear, });}
About
Zig library for reading and writing different image formats