Shape and Texture Based Plant Leaf Classification - Advanced Concepts for Intelligent Vision Systems - page 350

Table 3. Results for the two methods

Contour Score Sobel Score

1 Agrifolia 0.94 0.27

2 Castaneifolia 0.64 0.88

3 Ellipsoidalis 0.47 0.77

4 Frainetto 0.56 0.56

5 Hispanica 0.37 0.43

6 Ilex 0.78 0.81

7 Robur 0.64 0.72

8 Turneri 0.38 0.72

9 Variabilis 0.40 0.76

10 1982 0.97 0.32

11 1995 1.00 0.75

12 1996 0.45 0.29

13 1998-523 0.77 1.00

14 1998-4292 1.00 0.48

15 2005 0.44 0.88

16 2008 1.00 0.93

17 F184 0.91 1.00

18 Passifloranono 0.77 0.33

Table 4. The confusion matrix for the final, incremental classification

012345678901234567
0 .000000000000000000
1 0 .800 .20000000000000
2 00 .6 .300000000000000
3 000 .100 .800000000000
4 0 .700 .3 .8000000000000
5 0000 .7 .600 .6000000000
6 000 .400 .600000000000
7 00000 .20 .5 .2000000000
8 00000 .40 .6 .600 .4000000
9 000000000 .0000000 .00
10 0000000000 .00000000
11 0 0.27 0 0 0 0.06 0 0.07 0.18 0 0 0.40 0 0 0 0 0 0
12 000000000000 .000000
13 0000000000000 .00000
14 000000000000 .10 .8000
15 000000000000000 .000
16 0000000000000000 .00
17 00000000000000000 .0

5 Incremental Classification

It seems that leaves cannot be suciently well classified based on the shape or
the texture alone, though good results may be achieved by using both of these
features. In order to limit the risks of failure and improve the recognition rate,
we will use an incremental classification method. Firstly, the calculation of the
inflection points is used to separate the lobed and unlobed leaves. The species
which are in the same category as the leaf being analysed are kept and the other
species are ignored.