Page 86 - Read Online

Bah et al. Intell Robot 2022;2(1):72­88 I http://dx.doi.org/10.20517/ir.2021.16 Page 80

Figure 3. Data distribution among the 7 emotions.

Figure 4. Data split into training, validation and testing sets.

testing with a personal image. Data augmentation improves and enhances training dataset’s image size and
quality via suitable techniques [37] . The problem of overfitting that is common from the lack of sufficient data
is reduced through data augmentation [38] . This research uses data augmentation to transform an image to its
original state and train the CNN architecture.

The data augmentation is done by applying the geometrical transformation by first creating a new set of the
horizontally flipped datasets, image rotation, shift, and zoom, among other transformation operations [37] , and
adjusting the brightness to create new images of the same face. The images are also normalized to make the
pixel values range from 0 to 1. The provided images are then ready to be used to train the model.