OpenCV 4 with Python Blueprints - Second Edition

Crafting an MLP for facial expression recognition

Analogous to Chapter 7, Learning to Recognize Traffic Signs, we will use the machine learning class that OpenCV provides, which is ml.ANN_MLP. Here are the steps to create and configure an MLP in OpenCV:

1. Instantiate an empty ANN_MLP object, like this:

    mlp = cv2.ml.ANN_MLP_create()

2. Set the network architecture—the first layer equal to the dimensionality of the data, and the last layer equal to the output size of 6 required for the number of possible emotions, as follows:

    mlp.setLayerSizes(np.array([20, 10, 6], dtype=np.uint8)

3. We set the training algorithm to backpropagation, and use the symmetric sigmoid function for activation, as we discussed in the previous sections, by running the ...