Orthogonal global-locally discriminant projection for plant leaf classification

A dimensional reduction method named orthogonal ‘global-locally’ discriminant projection (OGLDP) was proposed for plant leaf classification in this paper. Given a set of data points in the ambient space, a weight matrix was firstly built which describes the relationship between the data points. Then the between-class scatter matrix and locally structure matrix were constructed by making use of the class information and locally information of the data, which can ‘push’ the within-class data points closer together, while simultaneously ‘pull’ the between-class data points even more far from each other. This character is advantage to data classification. Finally, the optimal objective function was set up, which was solved by Lagrange multiplication. The experiment results of plant leaf classification show that OGLDP is effective and feasible.