A two-stage approach for leaf vein extraction

Living plant recognition is a promising but challenging task in the fields of pattern recognition and computer vision. As an inherent trait, the leaf vein definitely contains the important information for plant species recognition despite of its complex modality. In this paper, an efficient two-stage approach is presented for leaf vein extraction. At the first stage, a preliminary segmentation based on the intensity histogram of the leaf image is carried out to estimate the rough regions of vein pixels. This is followed at the second stage by a fine checking using a trained artificial neural network (ANN) classifier. Ten features distilled from a window centered at the pixel are used as the input to train the ANN classifier. Compared with conventional edge detection methods, experimental results show that the proposed method is capable of extracting more precise venation modality of the leaf for the subsequent leaf recognition.