This project uses a trained convolutional neural network (CNN). A convolutional network consists of two parts: feature learning and classification.
Humans recognize objects by looking for specific features in our field of view. When we look at an object and see a license plate, four wheels, and a big metal box, we are probably looking at a car.
The feature extraction part tries to recreate this process by using kernels. Kernels are tools in computer vision that try to highlight specific features in an image. For example, this Sobel kernel is used for edge detection:
CNNs themselves learn many such kernels to recognize different features. After this process, the classification part of the network uses these features and learns to classify them.