On Statistical Pattern Recognition in Independent Component Analysis Mixture Modelling
A natural evolution of statistical signal processing, in connection with the progressive increase in computational power, has been exploiting higher-order information. Thus, high-order spectral analysis and nonlinear adaptive filtering have received the attention of many researchers. One of the most successful techniques for non-linear processing of data with complex non-Gaussian distributions is the independent component analysis mixture modelling (ICAMM). This thesis defines a novel formalism for pattern recognition and classification based on ICAMM, which unifies a certain number of pattern recognition tasks allowing generalization. The versatile and powerful framework developed in this work can deal with data obtained from quite different areas, such as image processing, impact-echo testing, cultural heritage, hypnograms analysis, web-mining and might therefore be employed to solve many different real-world problems.
Nominated as an outstanding PhD theses by the Polytechnic University of ValenciaPresent an excellent state-of-the-art literature review of the main applied theoretical foundations of statistical pattern recognition Gives new insights into independent component analysis (ICA) and independent component analysis mixture modelling (ICAMM) research in the context of statistical pattern recognition Defines a novel general framework in statistical pattern recognition based on independent component analysis mixture modeling