BRIEF: Fast, Compact Binary Feature Descriptors

Why it exists

Traditional feature descriptors like SURF were often too computationally expensive for real-time applications on low-power hardware. BRIEF was created to provide a feature descriptor that is extremely fast to compute and match, enabling real-time computer vision tasks on devices like mobile phones.

The mental model

Instead of describing an image patch with a detailed vector of floating-point numbers, BRIEF creates a simple binary string—a compact fingerprint. It's generated by performing a series of simple "is pixel A brighter than pixel B?" tests within the image patch around a keypoint. This binary representation makes comparison incredibly cheap.

How it works

After a keypoint detector finds an interesting point, BRIEF takes a patch of pixels around it. It then performs a fixed set of intensity comparisons on pairs of pixels within that patch. For each pair, if the first pixel is brighter than the second, it appends a '1' to a bitstring; otherwise, it appends a '0'. The result is a binary vector. To compare two BRIEF descriptors, you calculate the Hamming distance between them—the number of positions at which the corresponding bits are different. This is just a bitwise XOR followed by a population count (counting the '1's), an operation that is extremely fast on modern CPUs with the POPCNT instruction.

When to use it

Use BRIEF when matching speed and computational efficiency are your primary concerns. It is ideal for real-time applications on resource-constrained devices, such as mobile phones or embedded systems. Its speed leaves significant CPU power available for subsequent tasks in the application pipeline. It is a general-purpose descriptor and can be combined with any keypoint detector.

When not to use it

Avoid the basic version of BRIEF when your application requires strong robustness to in-plane rotation. While it handles many photometric and geometric transformations, it is not inherently rotation-invariant like SURF or SIFT. If you need to match objects that can appear at any orientation, you may need a more robust (and slower) descriptor or a variant of BRIEF designed to handle rotation.

One canonical example

In performance tests, BRIEF proved to be almost two orders of magnitude faster to compute than the U-SURF-64 descriptor on a single CPU core. When matching 1000 points between image pairs from the standard "Wall" and "Fountain" datasets, its recognition accuracy was shown to be comparable to SURF, demonstrating its ability to be both fast and effective.