A Novel Method for Leaf Recognition Based on D-LLE and Polar Coordinate Feature Extraction

Abstract

By extracting low-level features under the rectangular coordinate system, traditional leaf recognition methods typically have properties such as high dimensionality of extracted features, high-computational requirement and weak generalization performance. Based on the manifold learning algorithm D-LLE, we proposed a novel leaf recognition method under the polar coordinate system. The method first extracts from the leaf images high-dimensional features associated with polar coordinate as the preprocessing. Consequently, D-LLE is harnessed to reduce the features' dimensionality. In the low-dimensional space, we use the nearest neighbor classifier to make final determination. Experimental results exhibit higher effectiveness and efficiency of our method compared with classical traditional methods.