Binary Descriptors

One problem with traditional descriptors is that they consist of a vector containing a certain number of values, normally floating-point values because of possible normalization; consequently, both extracting this vector and performing the comparison stage require substantial computational time. One of the most promising alternatives is to extract a binary vector as the descriptor. The binary vector occupies less memory, and comparison only requires computing the Hamming distance between the corresponding binary strings. The Hamming distance can be computed very efficiently by taking the XOR of the binary strings and counting the set bits (POPCOUNT).