Recognition system for wooded species of Canary Laurisilva from its contour using kernel of Fisher

In this work it has been developed a novel system of automatic recognition for trees leaves, based on a hybrid classification method by means of a Support Vector Machine (SVM), using the Fisher kernel, and calculated from a Hidden Markov Model (HMM). An angle sequence has been used as an element of parameterization, extracted from the leaves contour, making such sequence invariable by rotation, movements and size. The application of these algorithms has been focused on the implementation of a recogniser of leaves of endemic trees from the Canary Islands, in particular, 16 different species of Canary Laurisilva. Obtained successful rates have been higher than 99.9%, according to the number of employed leaves in the training process. Key-Words: Pattern Recognition, Artificial Intelligent, contour detection, parameterization, classification, HMM, kernel of Fisher and SVM.