How to convert image to rgba8888 byte array?

[BenMcLean]

I don't know where support questions go so I'm posting here ... let me know if there is a better place.

I've got SixLabors.ImageSharp 1.0.4 from NuGet.

In my program, the user submits a .png file.

I want to get the height, width and a byte array in the format of rgba8888, where every four bytes represents one pixel and only pixels are in the array (no header data, no footer data, no metadata, just pixels) so that if you divide the array size by 4 * height then you'd get width and if you divide the array size by 4 * width then you'd get height.

There are lots of answers on the Internet for how to do this, but they are all for old versions of ImageSharp and do not work on the current version.

Even the documentation does not actually say directly how to do this. It has something about doing things one way with a normal image and another way with a multi-megapixel image and has you converting to a data type called Rgba32 without specifying how to get a byte array from there.

I want a byte array because I want to write code that is agnostic to what image library is being used and will only include ImageSharp in my test project.

[antonfirsov]

I'm assuming that rgba8888 is equivalent of our Rgba32 pixel type in my answer.

If your image is not insanely huge (above 130 Megapixels ~ 11000 x 11000 pixels), the TryGetSinglePixelSpan method should succeed and you should be able to cast the Span<Rgba32> to Span<byte>, and eventually copy it to a byte[] just like the docs suggest.

Note that these API-s are about to go in ImageSharp 2.0, because of the necessary changes of the underlying buffer management. There will be image.CopyPixelDataTo methods instead. See #1739 for more details.

I want a byte array because I want to write code that is agnostic to what image library is being used and will only include ImageSharp in my test project.

We are intentionally lacking an image.ToByteArray() and similar API-s, because the large extra allocations hurt performance a lot as applications scale. If performance matters, I really not recommend to go with the byte[] approach. Instead, you can define your own abstraction layer with an image base class that defines capabilities to enumerate pixel data, though I admit this is not a trivial job.

[BenMcLean]

Though if extra allocations are OK in your use case,

This is 1990s style retro pixel art for video games. Tiny images sizes. I think I'm good.

you should be fine with a method that extracts byte[] from Image.

Assuming I got ImageSharp 2.0, what would that method look like?

Right now, when I've got a byte array of a picture and I want to see what it looks like, I just run this to see the output at the end of my program:

SixLabors.ImageSharp.Image.LoadPixelData<SixLabors.ImageSharp.PixelFormats.Rgba32>(
  byteArray,//data
  width,//width
  height//height
  ).SaveAsPng("output.png");

That is literally the only job ImageSharp currently has and all I need it to do is the reverse operation of this.

[BenMcLean]

I don't see why is this the case.

Sorry, I guess I'm having a bad day. I shouldn't have said "from scratch" since yeah, I could probably do the abstraction layer thing.

Am just feeling really irritated that I'm having so much trouble with a task that should be so simple, even if inefficient.

[antonfirsov]

Assuming I got ImageSharp 2.0, what would that method look like?

Use the CopyPixelDataTo method I mentioned. T[] is implicitly convertible to Spans<T>.

byte[] data = new byte[sizeof(Rgba32) * image.Width * image.Height];
image.CopyPixelDataTo(data);

[ghost]

You can export to bytes like this (although starting from version 2.0 of the ImageSharp, it is now not possible to do it
in parallel, and it is a huge drawback - i hope the authors will create a multythreaded version of ProcessPixelRows method "a pixelAccessor per thread" - now it is single threaded)

private static byte[] ToRGB24(Image<Rgb24> image)
     {
         int pixelSize = GetPixelSize(ColorDepths.Bits8, 3);
         int height = image.Height;
         int width = image.Width;
         int rowSize = width * pixelSize;
         byte[] result = new byte[rowSize * height];

         image.ProcessPixelRows(pixelAccessor =>
         {
             for (int y = 0; y < pixelAccessor.Height; y++)
             {
                 Span<Rgb24> rgbs = pixelAccessor.GetRowSpan(y);

                 Span<byte> resultSpan = result;
                 Span<byte> rowSpan = MemoryMarshal.AsBytes(rgbs);
                 Span<byte> outputSpan = resultSpan.Slice(y * rowSpan.Length, rowSpan.Length);
                 rowSpan.CopyTo(outputSpan);
             }
         });

         return result;
     }

[ghost]

P.S. - although image.CopyPixelDataTo(result); looks like more efficient

[JimBobSquarePants]

Hey @BenMcLean do you have a Six Labors license? Because this all reads a lot like you’re asking for support.

[BenMcLean]

Yeah, I guess it would be a support request, but for non-commercial individual hobbyist open source development: not for a company. Didn't know about the licensing thing when I posted this. I guess that explains why there's not really much of a community forum or anything because the business model is to get people to buy the support licenses. Sorry, I want to respect that and don't mean to be a freeloader: was just trying to get a quick simple function out of the most obvious FOSS thing I found for it on Google.

[JimBobSquarePants]

No worries. #1941 (reply in thread) is the answer you need anyway.