A new algorithm for location recognition of Iranian car number plates, based on RGB color model and geometrical figures

Positioning license plates has an important role in vehicle identification and using unique identifier for vehicles is a challenging issue in the fields of traffic optimization, statistical analysis, and criminal investigations. Despite unique characteristics of license plates for Iranian and some European vehicles, more research and utilization of simple and cost efficient approaches for positioning license plates of Iranian vehicles in color images are required. Genetic algorithm was used for calculating optimal threshold value in this study. Also, time consuming transforms such as Fourier, Hough, Wavelet, and color space conversion were avoided. First in the proposed method, with the separation of blue channel from the color image and differentiating it from gray level and thresholding with the optimal threshold, binary image was generated. Then candidate area was determined using geometrical properties such as area, and length to width ratio. Finally, license plate length was calculated with the obtained width and standard license plate ratio. The proposed method generated the least candidate area and is very flexible such that it performed identical for Iranian vehicle license plates that contain different background colors and also for images with different scales. The results of the analysis performed in MATLAB environment with dataset composed of 150 images with the standard size of 480*640 pixels and different ambient light conditions and direct imaging of Iranian vehicles with national license plates, validated the 96.66% efficiency and accuracy of the proposed method. Evaluation of the proposed algorithm on a dataset with 80 images with 420*680 pixels size that were captured by highway speed cameras demonstrated an accuracy of 87.5%.