Shape and Texture Based Plant Leaf Classification - Advanced Concepts for Intelligent Vision Systems - page 351

Secondly, a classification using only the contour signature method is performed
(the shape of the leaf being the most important feature for classification). Leaves
for which the distance between their contour signatures and those of the leaf be-
ing classified are greater than some threshold are removed. The same procedure
is then performed on the remaining leaves using the texture histograms.
For the final remaining leaves, the distances between both contour signature
and the texture histogram are combined, and the leaf is classified as the same
species as the closest of these. The results forthisareshownintable4.Theoverall
classification rate is 81.1%, a clear improvement over the separate methods.
6Conluon
In this work, an ecient classification framework was proposed to classify a
dataset of 18 species of leaves.
Firstly, a classification based on the shape of the leaf is described. Two contour
signatures are calculated based on the distance and angle of contour points from
the leaf's centre. This operation is done for every leaf of the dataset and the
dissimilarties between the graphs are calculated using the Jeffrey distance. This
classification, called the contour signature method, presents quite good results.
Further improvement is made by the separation of the lobed leaves from the
unlobed leaves by the calculation of the signature's inflection points.
Secondly, a classification using the Sobel operator is used in order to capture
the dissimilarities of the macro-texture of the leaves. A histogram is formed from
the orientation and magnitude of the edge gradients. Finally, a method com-
bining the lobe differentiation, the shaped-based and the texture-based method
through the use of probability density functions is implemented. The incremental
process is intended to extract the most potential from each individual method.
The results show that 10 species out of 18 are successfully classified with a clas-
sification rate greater than 85% and 4 with one of more than 75%. The overall
classification rate was 81.1%.
The identification of the leaves is a dicult problem because there is often
high intra-species variability, and low inter-species variation. Nevertheless, the
approach adopted in this work demonstrates the classification of leaves using a
combination of relatively simple methods is a valid and promising approach.
References
1. Arivazhagan, S., Ganesan, L., Priyal, S.P.: Texture classification using Gabor
wavelets based rotation invariant features. Pattern Recognition Letters 27, 1976-
1982 (2006)
2. Casanova, D., de Mesquita Sa Jr., J.J., Bruno, O.M.: Plant leaf identification using
Gabor wavelets. International Journal Of Imaging Systems And Technology 19,
236-243 (2009)
3. Du, J.X., Wang, X.F., Zhang, G.J.: Leaf shape based plant species recognition.
Applied Mathematics and Computation 185, 883-893 (2007)