Description

Image-based approaches are nowadays considered to be one of the most promising solution to help bridging the botanical taxonomic gap, as discussed for instance. We therefore see an increasing interest in this trans-disciplinary challenge in the multimedia community. Beyond the raw identification performances achievable by state-of-the-art computer vision algorithms, the visual search approach offers much more efficient and interactive ways of browsing large floras than standard field guides or online web catalogs. Most existing systems for image-based plant identification do not consider large numbers of plant species and pay less attention on computational efficiency, and the state-of-the-art CBIR solutions are still insufficient for supporting large-scale plant species identification. In this project the system proposed a hierarchical multi-task structural learning algorithm is developed to support large-scale plant species identification. A bottom-up approach is developed to achieve hierarchical learning of more discriminative tree classifiers over the visual tree: For a given parent node, it contains a set of sibling coarse-grained categories of plant species or sibling fine-grained plant species, a multi-task structural learning algorithm is developed to train their inter-related classifiers jointly to enhance their discrimination power. A hierarchical multi-task structural learning algorithm is developed to train more discriminative tree classifiers over the visual tree and identify new plant species effectively. It is worth emphasizing that our visual tree has provided a good environment to identify the inter-related learning tasks automatically. The proposed algorithms can achieve very competitive results on both the identification accuracy and the computational efficiency as compared with other existing approaches.