Stereo Bare Footprint Biometrics Recognition Based On Curved Surface Features

Recently, with the rapid development of information technique, personal identification has become one of the hot researching points in the information processing field. Footprint is also a type of important trace information belonging to biometrics. Although footprint verification is applied to the domains such as criminal detection for a long time, in which a certain expert experiences are accumulated, but the domestic and foreign research on footprint biometrics is far from sufficiency for the reason that the footprint's forming condition is variable, and lot of difficulty should be faced in data acquisition and measurement, characteristic description and recognition.Researching on stereo bare footprint depth data, along with utilizing the pattern recognition technologies, this thesis addresses stereo bare footprint biometrics recognition. The main contributions are summarized as follows:1. A stereo bare footprints recognition method using two feature extraction methods, principal component analysis (PCA) and Fisher linear discriminant (FLD) analysis, is constructed. The method is efficient in running speed, and the footprint verification result achieves a 0.78 correct classification rate.2. A stereo bare footprints recognition method based on feature selection method is constructed, which firstly uses three different feature selection methods to reduce the feature dimensions. Then the method uses bayes decision rule to classify the footprints in the predigested model with the reduced feature dimensions. The footprint verification result achieves a 0.83 correct classification rate. But it runs slower than PCA and FLD method.3. A stereo bare footprints recognition method based on neural network ensemble is proposed. The method advances the capability of neural network ensemble by combining the training data disturbance and inputing features disturbance. The footprint verification result achieves a 0.81 correct classification rate with this method.4. A comparability determination algorithm of curved surface based on gauss curvature and normal vector is proposed. This method characterizes the associated distribution of gauss curvatures and normal vectors of the curved surface, which reflects the bending strain and the change tendency of the curved surface. It is invariant to translation and rotation, and adaptable for any rigidity 3D object. The experiments on some simple 3D objects and stereo bare footprints indicate that our method is valid.