Tuesday, 30 July 2013

NEURAL-NETWORKS IN MULTI-MODAL BIOMETRICS

Dimensionality lessening methods convert the data in the high-dimensional space to a space of fewer magnitude. The data transformation may be linear, as in principal components analysis, or non-linear. Many biometric spaces, such as facial biometrics, comprised of a large number of facial appearance which causes difficulties in scholarship and recognition process.

One of the main linear techniques for dimensionality reduction is a Principal Components Analysis (PCA), which performs a linear mapping of the data to a lower dimensional space in such a way, that the individuality of the data in that low-dimensional space is maximized. However, the resulting dimensions might not be always efficient for biometric applications according to ambiguous subspaces problem. In this section, a survey on dimensionality reduction and methods of choosing proper feature subspaces has been provided. The argument to the use subspace-clustering dimensionality reduction is made based on wide-ranging survey.
To validate the network and dimensionality lessening method, we have to explain to overall model of multi-biometric structure. A multi biometric system uses multiple sensors for data acquisition. This allows capturing multiple samples of a single biometric trait (called multi-sample biometrics) and/or samples of multiple biometric traits (called multi source or multi-modal biometrics). This also allows the system to enroll and validate a user who does not acquire a specific biometric identifier.
One of the main problems faced by biometric community is the biometric system dependability and performance. Thus, the first goal is to find the principal mechanism of the distribution of biometric features, or the eigenvectors of the covariance matrix of the set of biometric images. These eigenvectors can be thought of as a set of features which together characterize the difference between biometric samples. As the primary biometric samples, we select fingerprints and face images, since they provide important unpredictability in quality and a large quantity of multi-dimensional vectors.
The suggested methodology was successfully tested on fingerprint matching problem. The minutia extraction method which consists of estimation of orientation field, ridge detection, and minutia detection was applied, with geometrical aspects of the new method and the Hopfield Neural Network which is used for identification process.
We now provide the experimental results to showcase advantages of chaotic neural network for fingerprint recognition in both accuracy and circumvention (resistance to errors). The test database contains both high and very low quality fingerprint images, thus if high level of accuracy is confirmed on those samples, the system possesses both precision and circumvention. The first task was to find the minutiae set based on the algorithm given in previous section. A binarized version of the input image is created next. The finding process is done based in the ridge with one pixel width, so the in the next process the thinning filter will be performed on the image.

No comments:

Post a Comment