Introduction - On Statistical Pattern Recognition in Independent Component Analysis Mixture Modelling - page 16

1.2 Scope
The scope of this work is to explore new insights of ICA research in the context of
statistical pattern recognition. Thus, new algorithms based on ICA and ICAMM
are designed for feature extraction, semi-supervised learning, unsupervised hier-
archical classification, sequential dynamic analysis, and modelling of particular
problems. The thesis will show a more general framework for ICAMM, integrating
the increase of performance and flexibility obtained by the following: non-
parametric modelling of the source distributions; the incorporation of prior
knowledge using semi-supervised learning; correction of non-linear dependencies
in classification; use of any ICA algorithm in the learning process; construction of
higher levels of classification from the ICAMM parameters; and modelling of
sequential dependence in the feature observation record. The extensions are
thoroughly explained and emphasized through appropriate and demonstrative
examples. The results come from a large set of both synthetic and real data. The
simulations and experiments are aimed at BSS and classification including com-
parisons of the performance and advantages of the proposed algorithms with
traditional methods. Processing of the application data is off-line, and quantifiable
evaluation is obtained for synthetic data where the underlying patterns and dis-
tributions are known.
In addition to the theoretical subjects, this work aims to demonstrate its suit-
ability and usefulness with practical applications of the developed techniques in
real-world problems. The research carried out in the thesis is framed within the
following projects of the Signal Processing Group at Polytechnic University of
Valencia:
• ''Non-Linear Mixture Processor with Application in Detection, Classification,
Filtering and Prediction'', supported by the Spanish Administration and the
FEDER Programme of the European Community under grant TEC 2008-02975.
• ''Advanced algorithms for the detection-classification of ultrasonic signals'',
supported by the Spanish Administration and the FEDER Programme of the
European Community under grant TEC 2005-01820.
• ''Re-engineering of the natural stone production chain through knowledge-based
processes, eco-innovation and new organisational paradigms—ISTONE'', sup-
ported by The Sixth Framework Programme, European Union and the Con-
struction Technology Institute of Valencia under grant 515762-2_IP.
• ''Integration of information and communication technologies in the ambit of
conservation and restoration of archaeological and ethnographic materials: non
destructive testing by ultrasounds'', supported by Polytechnic University of
Valencia under grant PPI-05-04-5626.
The projects listed above were concerned with developing the following
hardware-software prototypes: (i) a quality control system based on impact-echo
testing [ 61 ] for the marble industry; and (ii) a ceramic characterization system for
archaeologists. In the first prototype, the impact-echo hardware technique is