I have a questions which is also mentioed in this answer and this one but I'm using the binary descriptor and need more informations:
I'm using BRAND descriptors as an input of LSH problem. The descriptor's size are 300*32 to 400*32, in which 32 is the length of the descriptor and images have 300 to 400 keypoints. The output of BRAND is a int matrix.
As this answer mentioned the inputs of LSH are vectors of D dimensions, it means every images will insert to the hash table as one vector of D dimension, now here are my question:
A. How can I convert the descriptor matrix of integers to a vector for input? Shall I just copy the matrix rows in sequence as a vector? OR Is it possible to convert each line of the descriptor, convert the items in binary and concatenate them to 256 bit binary and have a vector of D dimension of them?
B. Is it nessesary to convert the integer values of the BRAND descriptor to binary digits, in case of using L1 norm, that as far as I understood is the same Hamming distance for binary vectors?
Thank you very much in advance.
