Figure 8 – uploaded by salem almarri

Fig 1.1 depicts a schematic of model of a single neuron.  The mathematical relationships describing the neuron model are given in equations 1.1

Figure 1.2: Result of affine transformation of external bias input on neuron k on output Ux.

Figure 1.3: Threshold function and a Sigmoid function for  smooth, continuous range of values between -1 and 1 to predict outputs probability ir

Figure 1.4: Feedforward network with a single layer of neurons.  layer. No computation is made by the input layer of source nodes.

Figure 1.5: Fully connected feedforward multi-layer network with one hidden layer and one output layer.  the network.

Figure 1.8: Transition pathway of data pyramid

Figure 1.9: Rosenblatt’s Perceptron

Figure 1.10: Signal-flow graph of a perceptron  induced local field or activation potential v (equation 1.10), which is then applied to a hard

Figure 2.1: Convolutional operation

Figure 2.2: Pooling operation  element along the column until the whole matrix is pooled.

Figure 2.3: Fully connected layer  end of the CNN architecture, or somewhere intermediate.  connected layer as input as shown in Fig 2.3. The fully connected layer is usually placed at the

Figure 2.4: Representation of sentiment analysis of deep convolutional neural network  something as negative, neutral or positive. This is illustrated in Fig 2.4.

Figure 3.1: Representation of facial action coding system

actions.  state, but a complex chain of events that once simulated would evoke feelings resulting ir

cube [23]. The three signal substances are dopamine, noradrenaline and serotonin.  This is a three-dimensional model (shown in Fig 3.3) with signal substances that form

Figure 5.1: Deep learning, machine learning, and artificial intelligence

Figure 5.2: Extracted features of 8 channels (labels) of layers 3, layer 8 and layer 14 as visualized by Deep Dream feature visualization technique

Figure 6.1: Image samples obtained from JAFFE database [1]

Figure 6.2: Detection precision analysis of Fast R-CCN (VGG-16) trained on JAFFE

Figure 6.4: Detection precision analysis of Fast R-CCN (VGG-16) trained on FER-2013

Table 6.9: RAVDESS database processing configuration  Table 6.10: Fast R-CNN training options using RAVDESS database

Figure 6.5: Samples of processed images from RAVDESS database [3].

VGG-16 network model is shown to be performing poorly when compared to VGG-19  Table 6.12: RAVDESS trained Fast R-CNN benchmarked against other published work

Figure 6.7: Average miss rate analysis on three different network models

Figure 6.8: An undertrained network model, trained with max epochs of 3 (left) and an overtrained network model trained with max epochs of 30. (right).

Figure 6.9: Detected and classified facial emotion using trained Fast R-CNN object detector (right), in comparison with a facial expression obtained from the training dataset (left).

Figure 6.10: Percentage of detected emotions expressed by 100 different participants  tector was more confident that it was a fearful expression rather than a surprised one.

Figure 6.11: Trained Fast R-CNN classifying 5 expressed emotions correctly.

Figure 6.12: Rapidly protoyped system platform for Fast R-CNN deployment