A new reliable approach for Persian license plate detection on colour images

In this paper, a new approach for car license plate detection will be presented. The proposed method is faster than previous methods as it avoids time-consuming image transformations such as Hough transform, Fourier transform, wavelet transform, etc. The suggested method is based on modified mathematical template matching and colour analysis to detect the location of the car's plate. This new method detects all plates in a matching cycle using avoidance dynamic templates such as classic template matching methods, neural network, PCI based and other artificial intelligence matching methods. A modified column strip search to find a standard geometrical template in new types of license plates in Iran and most European countries has been proposed. Some advantages of this method are: high-speed process/ low response time, ability to setup and run on microprocessors and process colour images without any resizing and converting. The suggested method can be used in automatic toll stations, tunnels, high ways, intelligent parking and traffic zone areas as a real-time license plate recognition system. Plate characters are recognized by a Support Vector Machine with a homogeneous polynomial kernel of degree five. The results of experiments based on images from speed control cameras on high ways demonstrate the performance, exactness, speed and reliability of this method. Finally, the proposed method boasts 96 percent performance detection rate.