The Math Inside Every JPEG Just Got Faster on Your Phone
Faster image math means quicker photo loads, snappier editing, and less battery drain.
Every photo you text, post, or scroll past has been squeezed down to a smaller file by a piece of math called the DCT (short for discrete cosine transform). It's the engine inside JPEG, the format used for most pictures online, and inside newer formats like JPEG XL. When that engine runs faster, photos open faster, camera apps feel snappier, video calls use less battery, and servers spend less money handling the billions of images uploaded daily. It is invisible plumbing, but almost everything you look at on a screen passes through it.
Two researchers, Antoine Moevus and Max Mignotte, wrote a new library called BRiDCT that does this math more efficiently. The key point: they didn't invent smarter arithmetic — the same number of calculations happens. Instead, they rearranged how those calculations flow through a computer chip's parallel lanes, its tiny high-speed memory slots, and its slower main memory. Think of it like a kitchen: same ingredients, same recipe, but the pots and cutting boards are laid out so the cook never walks across the room. On one Apple M3 Max chip, this version clocked in faster than Apple's own Accelerate framework, the widely used FFTW and Ooura libraries, and the JPEG XL and JPEG-turbo image tools. Through the popular Python programming tool NumPy, it also beat OpenCV in all 130 tested cases, including rectangular images and squares up to 1024x1024 pixels.
The catch is scale and reach. This is a technical report, not a finished product, and it was measured mainly on one Apple laptop chip using ARM NEON instructions, so results elsewhere may differ. The authors also admit larger images hit size-dependent slowdowns, and their own diagnostics surfaced overflow issues at extreme values — meaning the speed gains don't come free at every image size. Nothing is broken, but nothing is plug-and-play yet either.
Still, this is a good reminder of where real-world speedups often come from. Not always a clever new formula, but careful engineering that respects how modern chips actually work. If tools like OpenCV, browsers, or photo apps adopt it, you'd never notice the change — except that everything would feel a little faster.
- BRiDCT speeds up the DCT, the math that shrinks every JPEG photo — without changing the math itself, just how the chip runs it.
- It beat Apple's Accelerate, FFTW, OpenCV, and JPEG XL tools in tests on an Apple M3 Max chip, including images up to 1024x1024 pixels.
- It's a research library, not shipping software, and gains were measured mainly on one chip — so don't expect your phone to speed up tomorrow.
Why It Matters
If adopted, faster image math means quicker photo loads, smoother camera apps, and less battery drain for everyone.