Stability analysis of classifiers for leaf recognition using shape features

Automation of plant species recognition is an important research problem that can have tremendous impact on saving valuable time of taxonomists and botanists and providing wide spread education about usage of flora in a community. Features based on shape of the leaves are one of the most commonly used features for automatic plant recognition as they are less affected by seasonal and environmental factors. This paper presents a detailed analysis of stability of four classifiers: Linear discriminant analysis, K-nearest neighbor, Treebagging and Support vector machine used for leaf recognition. The analysis shows that for automatic leaf recognition, Support vector machine based classifier is most stable with respect to quality of training images but it needs comparatively larger training set. The number of training images and the number of classes have varying impact on results of different classifiers.