Apostolos Ifantis

SUPPORT VECTOR IDENTIFICATION OF SEISMIC ELECTRIC SIGNALS

Traditional pattern recognition approaches usually generalize poorly on difficult tasks as the problem of identification of the Seismic Electric Signals (SES) electrotelluric precursors for earthquake prediction. This work demonstrates that the Support Vector Machine (SVM) can perform well on this application. The a priori knowledge consists of a set of VAN rules for SES signal detection. The SVM extracts implicitly these rules from properly preprocessed features and obtains generalization performance founded upon a robust mathematical basis. The potentiality of obtaining generalization potential even in feature spaces of high dimensionality bypasses the problems due to overtraining of the conventional machine learning architectures. The paper considers the optimization of the generalization performance of the SVM. The results indicate that the SVM outperforms many alternative computational intelligence models for the task of SES pattern recognition.

Color image enhancement using local density function estimators

In this work two combinations of potential and Euclidean-based functions that lead to edge enhancement of color images are proposed. Potential functions are calculated at each pixels neighbourhood, embodying spatial information and are used as local density estimators for the image. Color similarity information is included in the Euclidean distances in the RGB color space.

Exploratory analysis of electrotelluric field data for earthquake prediction

In this work recordings of electrotelluric field data are examined in order to detect earthquake precursor signals. Three different types of electric signals related to short-medium and long time variations of the Earths electric field are detected and analysed. These measurements are part of a larger multiparameter monitoring experiment performed in NW Greece over a period of 18 months.

A spectral distribution approach to fingerprint verification

A generic, transform-domain image classification method is presented and applied to the fingerprint verification problem. At first, the image is decomposed by a bank of Gabor filters and, at every pixel, its spectral information is extracted in vectorial form. In order to reduce redundancy, a neural-based vector quantizer is used to select representative samples that encode the multivariable fingerprint spectral distribution. Similarity between image distributions, utilized as a distance measure by the classification task, is then assessed in pairwise form by means of a non-parametric statistical test between the corresponding code-vectors. The presented multi-scale vectorial representation allows the inclusion of higher order dependencies among image pixels that describe in a unique way individual features of fingerprint images.

The nonlinear predictability of the electrotelluric field variations data analyzed with support vector machines as an earthquake precursor.

This work investigates the nonlinear predictability of the Electro Telluric Field (ETF) variations data in order to develop new intelligent tools for the difficult task of earthquake prediction. Support Vector Machines trained on a signal window have been used to predict the next sample. We observe a significant increase at this short-term unpredictability of the ETF signal at about two weeks time period before the major earthquakes that took place in regions near the recording devices. The unpredictability increase can be attributed to a quick time variation of the dynamics that produce the ETF signal due to the earthquake generation process. Thus, this increase can be taken into advantage for signaling for an increased possibility of a large earthquake within the next few days in the neighboring region of the recording station.

On-line signature recognition via fusion of dynamic features into dissimilarity space

In this paper a method for on-line signature recognition that combines dynamic features, fused into dissimilarity space, with a powerful collaborative sparse representation-based classification scheme is proposed. Dissimilarity vectors are formed in two stages. Initially, a number of informative dynamic features are extracted and stored in sequences. Afterwards, pairwise dissimilarities among feature sequences, computed using the DTW algorithm, are used to construct the new representation. Based on collaborative sparse representation principles, a new embedding space is defined where classification can be implemented efficiently. According to this scheme, signatures are represented in terms of their position inside the data structure, resulting in high-level performance without utilizing optimal feature selection procedures. The proposed framework has been evaluated using the SUSIG and the SVC2004 on-line signature databases.

Extraction and selection of geoelectrical data features

Extraction and selection of proper signal characteristics is a significant step affecting the success of any subsequent data analysis. We describe a process used to extract features from a 6-year, single-channel Long Term Geoelectric Potential difference (LTGP) signal, recorded in 1998–2003 at Western Greece. Features are extracted from consecutive segments of the signal, and evaluated to identify those possibly correlated with the seismic activity of the region. Evaluation is aided by pattern recognition techniques, and uses information from all seismic events of medium or larger magnitude occurring in the region. Initial results indicate that the approach may help reveal signal features whose properties deserve further investigation.

Third-order Spectral Analysis of Geolectrical Signals

A third-order spectral analysis for the detection of nonlinearities in geoelectric signals is presented in this work. Specifically, the bispectrum and bicoherence function are employed for the case of the geoelectrical signal acquired over the period of 1993-2002 in the area of Patras, Greece. This is an area of intense seismic activity and the possible relation between this activity and the non-linear mechanism of the geoelectrical signal is explored

Classification of multitemporal SAR images for the broader area of Athens

A comparison of different classification approaches for multitemporal SAR images data sets is provided in this work. The aim is to assess the performance of estimators of the backscatter temporal variability in terms of classification accuracy for a typical four-class problem. Different approaches in forming an appropriate feature vector are discussed and compared with multichannel classifiers like the fuzzy k-means. Finally, a classifier that employs a feature fusion step based on principal components analysis is proven promising since it provides increased classification accuracy and reduced computational complexity.