Reconstructing images from 1-bit quantized local descriptors

The paper "From Bits to Images:Inversion of Local Binary Descriptors", by Emmanuel D'Angelo, Alexandre Alahi, Pierre Vandergheynst and Laurent Jacques shows it is possible to reconstruct accurate images from the information contained in binary descriptors used in computer vision applications. The authors formulated the reconstruction as an ill-posed inverse problem, regularized by the sparsity of the wavelet representation of natural images. Since local descriptors are 1-bit quantized, the paper also borrows ideas from 1-bit Compressive Sensing to formulate an efficient algorithm. The results shed new light on the information contained in popular binary descriptors such as BRIEF and FREAK. But it also raises privacy concerns, since these descriptors are used in a lot of mobile augmented reality applications.