Page 172 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 6 Plant leaf disease classification based on feature selection 161

and preserves other hue and saturation values. Afterward, the
intensity is divided by separator into two subparameters which
are high and low groups. This is done by the golden section
search method shown in the following equation:

g ¼ fgðiÞji > g g; g lo ¼ fgðjÞj j g g (6.1)
hi
m
m
where g hi and g lo are intensity high and low groups, respectively,
and g m is a trial threshold intensity value which is defined to
divide the image into two subimages. After obtaining estimates
of the two subparameters of intensity, a combination of them is
performed to achieve the enhanced intensity. The enhanced
intensity is calculated by the following equation:

g enhance ðiÞ ¼ g þ ðg g Þ cðiÞ (6.2)
lo
hi
lo
where c(i) is the cumulative intensity of i pixels. To ensure that
brightness error is minimum, the values of calculated mean
brightness and input brightness are compared. In other words,
iteration of this process is performed until getting an optimal
value of enhanced intensity. Eventually, enhanced intensity and
other initial hue and saturation values are combined and con-
verted back to RGB color channel to give output image. The
contrast enhancement effect is illustrated in Fig. 6.5.

3.3 Convolutional neural network

A ConvNet/CNN is a DL algorithm, which can take in an input
image, assign importance (learnable weights and biases) to
various aspects/objects in the image, and differentiate one from

Figure 6.5 Contrast enhancement effect: (A) before and (B) after.