Image analysis for automatic classification of lichens by taxonomic identification of the type of thallus growth

Abstract

Recently, the study of lichens has acquired relevance by their potential for production of medicines as well as bioindicators of air quality and ecosystem health. However, computational studies for automatic classification of lichens are few. These investigations depend on availability of large volumes of images for the training process, a scenario quite far from reality. Therefore, it is crucial to identify image descriptors for improving classifier performance, under a limited number of samples. This article introduces a novel method to automatically classify lichens using a descriptor, robust to scale, rotation, and illumination variations, and based on analysis of textures and color. By applying a support vector machine (SVM) classifier, an average F1 score of 99.3% is achieved when classifying lichens in the three categories: crustose, fruticose, and foliose. The method supports the work of taxonomists and facilitates inexperienced users to identify and characterize lichens.