A part of the project is dedicated for the security of biometric data. For this purpose we’ve implemented the biohash algorithm to encrypt palm vein and palm print data. First biometric data is processed with CMF to obtain the feature vectors that represent the palm print or vein structure. From this information we create a data vector (of size n) that consists of feature vector descriptors.
We also create a pseudo random matrix (of size nxn) using a token as a seed. Then we calculate the inner product of the data vector and every row of the generated random matrix. Those values are the thresholded (every row has it’s own threshold value). The resulting series of ones and zeroes is called biocode, that is then used to identify a persons biometric sample.
Two adjustments were made to the original method. The first one is adding additional higher level data to the data vector that previously consisted only of the raw vectors. The new data is the information about the most positions of the most intensive vectors found in the biometric sample. It grants the same result even if the order of the most intensive vectors changes.
The second improvement is the biocode comparison approach improvement by adding additional information to the biocode. The added data is statistical data about the resulting biocodes. And the data is calculated for every person. The statistics tells which bits in the biocode are more stable and change less. This allows us to discard those bits that are basicly changing at random and add more emphasis to those bits that change less frequently. The data is a list of indices that order the biocode bits by their stability. And comparing two biocodes, the resulting error is calculated by using a weight function.
This algorithm was tested on palm print and vein image database from 50 persons, where 5 images from each person are taken in near infrared (NIR) spectrum to capture palm vein image and 5 images that are taken in visible light spectrum to capture palm print image. Total of 500 images, 2 images for training and 3 for testing from each person. The obtained results and comparison to the raw biometric data comparison is shown in the table below.
