Robust Chinese License Plate Recognition System Using Deep Learning Techniques

This research work proposes an efficient,reliable,and robust LPR system which is totally based on machine learning algorithms.We have introduced a new method for segmenting Chinese license plates in which the license plate segmentation is performed using powerful deep learning techniques instead of traditional digital image processing techniques.Firstly,input license plate image is preprocessed using traditional digital image processing techniques:input image is converted into gray scale,and then skew detection and correction is performed.Secondly,license plate is segmented and characters are separated using a well-trained convolutional neural network(CNN)so that each character will be stored in its own image.Those characters can be later recognized and classified using any character recognition module.The proposed license plate segmentation method is straight forward,less complex,and can be considered as a good alternative for some traditional digital image processing license plate segmentation methods.Also the main idea of proposed segmentation method has good extensibility so that it can be extended to any kind and format of license plates easily.In addition to plate segmentation,character recognition and classification is considered one of the most important parts of current LPR systems.Because of low recognition quality and poor robustness of traditional character recognition techniques,those techniques were gradually replaced by powerful deep learning modules such as convolutional neural networks.Convolutional neural networks show satisfying ability in character recognition and outperform most of other available models.Since Chinese license plates contain both Chinese and alphanumeric characters,a robust,powerful,and efficient CNN is required to accomplish character recognition task efficiently.In this research work,we have proposed an efficient CNN model based on Darknet architecture to perform character recognition.Through convolutional and maxpooling layers,features of input character images will be extracted and then sent to softmax layer for classification.To avoid overfitting problem in the training process,the dropout regularization technique is adopted.We have used a dataset of 12000 character images for training and testing our whole LPR system.The experimental results show satisfactory performance and eventually achieve test accuracy of 97.45%.