Storing and comparing biometric information
In general, we use biometrics in computer applications for authentication. Lets get 2 examples of fingerprints and face recognition.
Where we store information for comparison. As an example, we cannot save the image and process it every time. So what methods do we use to store / determine the similarity in such cases? Are there special algorithms for this purpose? (Example: return roughly the same value for a specific person's fingerprint every time)
source to share
Most AI techniques don't work with raw data such as images. As a rule, they work with a feature vector: preferably a compact and intelligent representation of the original data. Typically, a characteristic vector contains a fixed number of numeric or nominal values (characteristics). For example, in face recognition, a common feature vector is a set of eigenvectors called Eigenface . I am not familiar with fingerprint recognition, but my guess is that the feature vectors used contain a set of numbers that somehow describe the observed fingerprint image patterns.
Typically, when training machine learning on a set of facial or fingerprint images, you compute the appropriate feature vectors for those images and store them in a database. The original images are no longer used. All subsequent processing is performed on the corresponding feature vectors.
To compare the new, invisible instance with the database of previously learned instances, the function vector of the new instance is computed and compared to the database of stored feature vectors. This can be done in a number of ways. One example that is commonly used in iris recognition is Hamming distance .
source to share
In the case of fingerprint analysis, I have heard of people using the arrangement of points of objects (bifurcations, etc.) to match parameters for a large polynomial, and then save the parameters to match when someone wants to explore the gallery. (The matching process seems to work, minimizing the derived error rate between the probe and gallery parameters.) I've never done this myself since I'm mainly working with irises, but it might be worth looking into it.
source to share
All biometric connectors work with processed data called templates. This data is taken from a static image or a model taken from a dynamic capture as previously discussed in TC. These templates are used to match the process and are unique data that you need to store. The images are only kept for listening or in criminal cases, when an expert must analyze the images and accept the final result.
We use 3 international standards for fingerprint templates: ISO 19497-2, ISO-378 and XYT. If you use either of the first two standards, binary data is usually 500 bytes long. XYT needs more space because it is a text file with position, angle and quality of all minutes, usually around 1 kb. You can see a sample extraction and mapping on the NIST website . Therefore, if you want more accurate and faster software, you need a commercial SDK.
For dealing with individuals, the free software is OpenCV.
source to share