Tree recognition for landscape using by combination of features of its leaf, flower and bark

In this paper, we give a tree recognition system for landscape based on combination of the component retrievals, leaf, flower and bark. Segmentation that divides an image into object and background is implemented by user. Feature extraction is to get features of shape, color, and texture, which it depends on the characteristics of each component. We used 32 features of Fourier descriptor for leaf, 72 features of HS color plane for flower, and 20 features of wavelet energy for bark respectively. A Euclidean distance between features is used as a measure for retrieval. We collected a dataset with 16 classes for landscape tree and recognize the species with the first sequence of results for retrieval, which is having a minimum distance with query. We can obtain 31%, 75%, 75% recognition rate of leaf, flower and bark image in our dataset respectively. For combination, we added the sequence of retrieval for each class. We obtain 84% recognition performance in combination of leaf and flower, 75% in combination of leaf and bark, and 100% in combination of bark and flower. We showed that a combination of the component retrievals can be inferred a better performance in recognition of tree for landscape.