Convolutional Neural Network (CNN) is one of the artificial neural network types, which imitates the connecting pattern of human brain cells. Among several types of artificial neural networks, CNN is used prominently more than the others on computer vision analysis (image or video data). For instances, CNN can identify the id written on the car's license plate, human individual's face, and even animal species in the given image/video data. Below image represents the ground structure of CNN.
CNN uses a system much like a multilayer perceptron. The layers of a CNN consist of an input layer, an output layer and a hidden layer that includes multiple convolutional layers, pooling layers, fully connected layers and normalization layers. However, as written above, the image is just the ground structure of CNN: there exists several structures even among CNN models! The reason is that the hidden layer can be customized. For example, the number of convolutional layers, size of pooling, length of fully connected layers, and activation function are free to choose.
Accordingly, a global computer vision competition named 'ImageNet Large Scale Visual Recognition Challenge(ILSVRC)' is annually being held. The following chart shows the error estimations of the CNN structures of annual winners in ILSVRC from 2010.
To sum up, CNN is the most appropriate artifical neural network model for image recognition. Even today, many researchers are trying hard to discover the wider range application of CNN and raise the network's accuracy.
