Extraction of number plate images based on image category classification using deep learning

Automated Number Plate Recognition is a technique involving image processing which is used to identify a vehicle by reading its number plate. A proposed method is capable of extracting the number plate region in the image captured from its rear end at various car distances. The system analyzes the input image and identifies the location of the number plate. The plate candidate region is extracted by using dilation morphology and scoring based on the height-width ratio and the number of connected components in the region. The candidate region images are classified in two classes, namely number plate and outlier images. As image feature extractor, pre-trained CNN (convolutional neural network) is used and as classifier, SVM (support vector machine) is used. The algorithm was implemented by programming language C++ in morphological image processing and MATLAB in pre-trained CNN and SVM parts. The paper shows experimental verification of the algorithm and test results for 126 car images.