Local discriminant bases representation and non-linear growth processing for species classification and age estimation of fish based on otolith images

In computer based fish identification from otolith images, recent techniques such as Curvature Scale Space (CCS) and Wavelets have been used to locate the position of singular features. On the other hand, automated fish ageing methods use Fourier Transform (FT), Principal Component Analysis (PCA) or Peak Based Representation (PBR) to enhance classifier performance. In this work we use the Local Discriminant Bases algorithm (LDB). In ageing applications, is necessary to employ an optimization method to demodulate growth effects before using the LDB. LDB can also be used for fish identification after filtering and resampling the otolith contour. The proposed methods have been tested with otolith contours from a web-base database and with images of cod otolith sections. In both cases results show that only a few bases are needed to represent class features accurately.