Car license plate detection by color processing and edge detection methods

One of the main processes in intelligent transportation systems is car license plate detection. The most important part of this  algorithm is detecting the blue part on the left side of license plates. After identifying the blue part (width) of a license plate, the  length of the license plate can be determined since its width is proportional to its length. A new algorithm is presented in this paper to simultaneously perform the thresholding operation on the color and edge of the images obtained after proper preprocessing. Since the blue part of a license plate is near its left edge, the common features of the two obtained images are merged. The merged image is then used in Hough transform to detect the license plate. After detecting the license plate and according to the constant ratio of the width of each character to the length of the license plate, the location of each character is determined and the value of the character is recognized using a multilayer perceptron (MLP) neural network. Since this is a cascaded method, it can benefit from the advantages of other methods in different areas, such as color processing, edge detection, and morphological operations. Moreover, because most images are color images, the proposed method effectively uses color features, while this usage is less  common in the literature. To evaluate the accuracy of the proposed method, some images are collected using a mobile phone camera or randomly from the Internet. A total of 288 images are used to assess the accuracy of the license plate detection, and 2200 images are used to measure the validity of the license plate character recognition. The accuracy of detecting license plates is 93.75% in this study.