Local feature extraction and its applications using a library of bases (Q2709031)

The author uses a minimum description length (MDL) principle to achieve simultaneous noise suppression (additive white Gaussian noise) and signal compression. Because of the MDL criterion, this algorithm does not require the user to specify any parameter or threshold values. Coifman's shift-denoise-average algorithm is applied with the MDL based algorithm to reduce noise and Gibbs-like phenomena around edges so that the residual error becomes closer to white Gaussian noise. An algorithm is derived for selecting a local discriminant basis (LDB) suitable for classification from a library and for the extraction of signal components from data consisting of signal and textured background. A local regression basis and LDB are applied to a real geophysical regression problem to extract useful features. The author introduces a library of non-orthogonal bases using the autocorrelation functions of wavelets for multiscale edge characterization and representation.NEWLINENEWLINEFor the entire collection see [Zbl 0955.43001].