Abstract:
|
The application of feature extraction methodologies
and the detection of patterns in sagitae otoliths, which are
calcified structures in the inner ear of teleostean fishes, has lead
to great knowledge of marine biology during the last decades in
order to manage and control its sustainability. A main limitation
of the use of statistical analysis and Fourier methods rely on
their incapacity to locate irregularities and explain them from a
more structural, or even physical, point of view. This matter has
been addressed recently by means of the Best-Basis paradigm
which combines robust description methods, such as the Wavelet
Transform, and the potential of statistical analysis in order to
fully automate the selection process of efficient features. This
paper is an attempt to readdress this paradigm towards this
goal and contrasts other standard tools used in the field of
otolith-based fish recognition. The proposed strategy involves the
estimation of class distributions, discriminant measures and the
search in the feature space. |