A vehicle license plate detection method using region and edge based methods

In this paper, a license plate detection system is implemented. For this purpose, we improve the contrast at possible locations where there might be a license plate, we propose a filtering method called ''region-based'' in order to smooth the uniform and background areas of an image, we apply the Sobel operator and morphological filtering to extract the vertical edges and the candidate regions respectively, and finally, we segment the plate region by considering some geometrical features. In fact the novelty and the strength of our license plate detection system is in applying the region-based filtering that decreases the run time and increases the accuracy in the two final stages: morphological filtering and using the geometrical features. The experimental results show that our proposed method achieves appropriate performance in different scenarios. We should mention that our system is reliable because the accuracy is above 92% in average for different scenarios and it is also practical because of the low computational cost. An intelligent transportation system (ITS) is an important tool for analyzing and controlling the moving vehicles in cities and highways [1] and in recent years much research on ITS has been carried out. Nowadays automatic vehicle license plate (VLP) recognition is a key ingredient for any ITS such as security control of restricted areas, traffic law enforcements, automatic payment of tolls on highways and parking management systems. In these examples, a camera captures the vehicle images and a computer processes the captured images, detects the car license plate from the input image and then reads the information on the license plate by applying various image processing and optical character recognition techniques. Generally , an automatic VLP recognition system is made up of four modules: image pre-processing, license plate detection, character segmentation and character recognition (Fig. 1). Among these four modules in Fig. 1, license plate detection is the most important and the most difficult task in any VLP recognition system because of images with low contrast, blurring and dirty plates. The most common solutions for VLP detection include analyzing the texture [1,2], edge extraction [3–6], morphological filtering [7,8], color feature [7,9], combining of edge statistics and color feature [10], combining of edge and morphology operations [11], Hough transform [12], neural networks [6], learning-based approaches [13], Gabor filtering [14] and Wave-let analysis [15]. An edge approach is normally simple and fast. Texture [1,2] and edge based methods [3–6] are widely used …