CERBERUS Project, Human Verification: a Biometric Approach

The final goal of the Cerberus project is to create a system which is an alternative for the traditional password based verification. Two individual biometric features are used for verification: the face and voice characteristics. The arguments for this solution are that these are unique features of each individual. The system is currently in a developing phase. It means that the main modules work separately in three parts, such as a face recognition, a voice pre-processing and a neural network test programs. The complete program, the Cerberus Security Verification Software will be based on these parts.
For the voice pre-processing LP (linear prediction) analysis is used for spectral analysis and pitch period, which emphasis the significant information. In the face recognition part a global detection method, the Randomized Hough transformation is used to determine the face position on the picture. After finding the face, the eyes and the mouth are detected. The classification test program is a backpropagation neural network implementation. The decision is based on the values of the output cells. If they are near to “1” the samples are acceptable, if there is a little difference, a new sample is required, and if there is a bigger difference the sample is rejected.