Introducing the RGBA bitmap file format

Ben Zotto
5 min readFeb 20, 2021

.rgba is the dumbest possible image interchange format, for your programming pleasure.

So simple it fits on the back of an envelope.

TL;DR

Join me in supporting this dumbest of all possible bitmap image formats to make writing little tools and programs with images easier for everyone. It’s called RGBA and it stores your RGBA data. Here’s some free code if you want some, but the whole spec is up there in the scribble. Have at it.

Why this revolutionary idea?! and why now?

It’s time for a bitmap image interchange format that you can read and write comprehensively with the barest minimum of code and zero edge cases. The RGBA format I am proposing here is the new, beautifully stupid standard that you didn’t know you needed until now. The actual spec is short enough to fit in a tweet, or on the back of an envelope.

If you’re a programmer dealing with image or bitmap data, you’ll be familiar with the idea of the raw bitmap in memory, which these days is almost always one big, linear, RGBA, 4-byte-per-pixel buffer. You’ll be equally familiar with all the annoyances of getting this data to and from a file, the network, etc.

Why? Because although the bitmap in memory is super easy to work with, image files are all complicated, designed around the computing constraints and multifarious graphics hardware of the 1990s (!). Those standards all require you to pull in complex libraries and frameworks to work with them.

If you want to keep your code self-contained and simple there aren’t a lot of options. PNG and GIF are nontrivial; JPEG will make you run screaming. Even Microsoft’s BMP format, which is the nearest common format to “raw,” is a rat’s nest of accretive complexity once you start trying to parse it. Here is a handy diagram illustrating the different pixel subformats that BMP supports:

DON’T TRY THIS AT HOME FOLKS .Microsoft BMP format variants, via Wikipedia.

Who the f — needs to encode and decode 4 bit-per-pixel indexed palette RLE data in this futuristic year 2021? NO ONE. Disks are big and cheap, networks are fast. RGBA is fine, even if you’re only using a few colors!

So in some sense, .rgba is a spiritual successor to the Windows BMP — a distillation of its very essence into the single and minimal form that the vast majority of those files were using anyway.

Here is the spec for the file format:

  • 32-bit “magic number to identify the file. This is the characters RGBA, in that order.
  • 32-bit width (in pixels). Big-endian.
  • 32-bit height (in pixels). Big-endian.
  • Pixel data: 4 bytes per pixel, one byte each R, G, B, A. Always in that order. The bitmap starts at the upper-left of the image and scans each line, all the way to the bottom. There is no padding anywhere. The total length of the pixel data will be thus equal to width * height * 4 , and the formula for accessing the first byte (red channel) of any pixel will be: pixel_data[(y * width * 4) + (x * 4)]

The two numeric values, for the dimensions, are in big-endian byte order, also known as network byte order, also known as the one that’s easier to look at in a hex editor or memory dump.

I am providing a handy C module to conveniently save and load if you want to use that, but the format is so simple that you can do it in lots of different ways.

F.A.Q.s

What is the file extension for these files?

.rgba of course.

Won’t these files be huge?

Lol maybe. But so is your disk and your Dropbox and your RAM so who cares. This is not meant to take the place of a judicious JPEG, obviously. Just an option to have when you’re building simple tooling.

Can I store the pixels in ARGB or BGRA order?

No, you can’t! It’s right there in the name. I mean, you can do anything you want, but the .rgba file will only be correct if it’s in actual RGBA (a.k.a. R8G8B8A8) order. This was a kind of hard call, because there are some pretty reasonable scenarios why you might find it convenient if the file data mapped exactly onto your slightly different e.g. texture layout.

But once you introduce the option to have other orderings, you immediately end up with half the code that reads the files being incorrect or incomplete, so they mostly work, but not always, and then you might as well just be using the BMP format.

The point is clarity and simplicity and portability. You may need to transform the data slightly when reading or writing it. But since you’re already reading a large and uncompressed image, is that so much extra work? :)

Where does the other image metadata go?

It doesn’t! You could, I suppose, append additional data in your preferred format after the pixel data. A reasonable enough decoder might probably just ignore it? But I neither encourage nor discourage this course of action.

How do I specify a color space? Don’t I need that?

I barely even understand what that gamut stuff means. Isn’t 16 million colors enough for you?? If you need to make an assumption about color space, assume sRGB (a.k.a “Standard RGB”) because that’s like 99.9% of bitmap data anyone cares about in the real world. (If you’re doing work that distinguishes color spaces, you’re probably looking for a different file format.)

My bitmaps don’t ever use transparency! Files are 33% too big!

Sigh. Yeah, it’s a fair question, and it’s the strongest case for allowing more than one kind of byte format. But simplicity wins here. Maybe in the next format. :)

Is the alpha channel premultiplied?

No. And if your image doesn’t have transparency, just use 0xFF for the alpha channel.

What about alignment?

Everything (the three header fields and every pixel) is naturally 32-bit aligned, and no padding anywhere is required (or allowed). This should suffice for the vast majority of needs.

(There are hypothetical cases where you might wish to load data directly into a buffer that benefits from 64-bit alignment. You can of course load the bitmap separately from the header fields and ensure your width is a multiple of two.)

How does .rgba beat the competition?

In simplicity of implementation, and speed of read/write. Any compressed format is more complex (PNG, GIF, or my god, JPEG). Windows BMP is the closest to allowing raw data but you still need to screw around with a bunch of semi-documented header fields and pick your bitmasks correctly. Any file format that needs a corpus of varying sample files to reasonably validate a decoder is not simple enough!

Who the hell are you?

Just some guy who’s sick of spending more time decoding image files than doing the actual projects I’m trying to do. You’re welcome.

--

--