Bayesian Approach in Kendall Shape Space for Plant Species Classification

Modelling computer vision problems with Riemannian manifolds yields excellent results given that the visual features of these maniflods have special structures that Euclidean space doesn’t capture. In this paper we propose an approach based on the Kendall manifold formalism and the Bayesian approach applied to a plant species classification problem. Kendall space is a quotient space that is provided with a Riemannian metric, which is more convenient when shapes differ only in translation, rotation and scale. However, an appropriate metric for shape classification task should not only suit certain invariance properties but also satisfy the different input properties. The non-linearity of Kendall space makes it difficult to apply common algorithms for classification. Thus, we propose to adapt the Bayes classifier to Kendalls representation using landmarks. Our main contribution consists in computing the parameters of the likelihood density functions through tangent spaces. Experimental results show that our approach is more accurate and effective.