Machine Learning Approach to Improve Flower Classification Using Multiple Feature Set

Abstract

Image based classification systems are achieving best performance using large image datasets and advanced classification techniques. Most of the flower classes have same shape, appearance or background information such as plant leaves and grass. So, flower image classification is still challenging task. The goal of this paper is to analyse the effect of multiple local features for flower image classification. Shape, texture and color features are extracted from the flower images in order to describing different aspects of flowers. The classification performance of the proposed method is also compared with state-of-the-art flower classification performances. Performance of the local feature descriptors such as SIFT, SURF, HSV, RGB and CTM in flower classification is also analysed. According to the performance of the local descriptors, the combined SURF + CTM gives better performance than other combination of features in the context of flower image classification.