Oak Leaf Classification: An Analysis of Features and Classifiers

Abstract

Automatic classification of trees from leaves is a popular computer vision/machine learning task and has important applications in monitoring of forest wealth. While the final aim is preparing an application, which is capable of visual signal processing and classification, in this paper we present a new oak leaf dataset and preliminary results for classification of 8 types of oak trees. The novelties include comparative analysis of a small set of hand-crafted geometric features and popularly used high-dimensional appearance features, such as Local Binary Patterns (LBP) and Histograms of Oriented Gradients (HOG). We further compare commonly used Support Vector Machines (SVM) classifier with a recently popular, fast and robust learner called Extreme Learning Machines (ELM). Results indicate that a small set of geometric features reach an accuracy of 75%, while high dimensional appearance features can boost the performance up to92%.