Due to the large difference between the same-class leaf images, many classical recognition methods do not satisfy the actual requirements of the plant leaf image recognition system. Based on maximum variance unfolding(MVU) and maximum variance projection( MVP), a supervised orthogonal MVU algorithm was presented and was applied to plant leaf image recognition. By the algorithm, the high-dimensionality data were mapped to an optimal low-dimensionality subspace where the different-class samples were located further away, while the same-class samples were located closer. The local geometry structure of the low dimension manifold of the original high dimensionality data was preserved. The experimental results on real plant leaf databases showed that the proposed method was effective and feasible for plant leaf recognition.
展开▼
