A Convolutional Neural Network, also known as CNN, is a class of neural networks specialized in processing images. The main features of CNN are the following:
• It typically has three layers: a convolutional layer, a pooling layer, and a fully connected layer.
• The convolution layer is the core building block of the CNN. It carries the main portion of the network’s computational load.
• Convolutional layers take as inputs tensors of shape (image_height, image_width, color_channels) and the output is a 3D tensor of shape (height, width, output channels).
• The output of a Convolutional layer is the result of the convolution of a moving filter, also known as kernel, applied to a certain neighborhood of nodes (which are usually the pixels of the image). Therefore, instead of learning a number equal to the input’s size of weights and bias for each hidden neuron, as in fully connected layers, the network has to learn only the values of the kernel.
• Filter’s parameters are filter’s size, number of filters in the convolution, strides of the convolution along the height and width, application or not of padding.
• The pooling layer involves sliding a two-dimensional filter over each channel and summarizing the features lying within the region covered by the filter. Pooling layers are used to reduce the dimensions of the feature maps, leading to a reduction in the number of parameters to learn, and the amount of computation performed in the network.
• The fully connected layer with softmax as activation function is used to make the classification of the images.

Lawrence, S. and Giles, C. L. and Tsoi, A. C. and Back, A. D., Face recognition: A convolutional neural-network approach, IEEE transactions on neural networks 8.1 (1997): 98-113.

Giulia Cademartori

University of Genoa