A First Course in FUZZY and NEURAL CONTROL

5.7. EXAMPLE 2: TRAINING A NEURAL NETWORK 187

Figure 5.12. Leaf sample and segmented image

Figure 5.13. Pepper sample and segmented image

Table 5.1 lists two sets of features that were selected to train backpropagation
neural networks with various topologies. The purpose of course was to determine
which set of features yielded the best recognition using an appropriate network
topology.

Table 5.1. Features used to train neural network

Feature set 1 Feature set 2

Area, Solidity, and Edif Area, Shape Factor, Solidity,

Convexity, Compactness,

Aspect Ratio, and Edif

The feature parameters listed in Table 5.1 are described by empirical rela-
tionships. They are primarily shape descriptors and are briefly described here for
completeness. The methods for obtaining these features are well documented
andareavailableintheMatlabimage processing toolbox. A list of several
other descriptors can be obtained from any image processing handbook.

Area:Since images are represented in terms of pixels, the area of an object is

the total number of pixels within the object.

Perimeter: The perimeter is the number of pixels along the boundary of the

object.