Abstract
Binary descriptors of image patches are increasingly pop- ular given that they require less storage and enable faster processing. This, however, comes at a price of lower recognition performances. To boost these performances, we pro ject the image patches to a more dis- criminative subspace, and threshold their coordinates to build our binary descriptor. However, applying complex pro jections to the patches is slow, which negates some of the advantages of binary descriptors. Hence, our key idea is to learn the discriminative pro jections so that they can be decomposed into a small number of simple filters for which the responses can be computed fast. We show that with as few as 32 bits per descriptor we outperform the state-of-the-art binary descriptors in terms of both accuracy and efficiency.