Rotary matching of edge features for leaf recognition

With advances in cloud computing technology, handheld computers and smartphones can now perform plant recognition by taking a photograph of a plant. This study proposes novel features to describe leaf edge variation. The Bayes theorem is used to calculate the maximal matching score for rotary matching. The Viterbi training algorithm is then applied to find the model parameters of rotary matching. The experimental results show that the top one of 13-tuple reaches 94.4% and the first two can also achieve 100% in the test set. The results have verified that the proposed features are invariant to translation, rotation and size.

► Features are extracted to describe variation of leaf edge.
► Viterbi training algorithm is applied for leaf modeling.
► Bayes theorem is proposed to calculate maximum matching score for model matching.
► The experimental results show that the top one of 13-tuple reaches 95.9%.