Automatic Recognition and Localization of Saudi License Plates

Introduction

Recognizing characters automatically in license plates has become a necessity in this age. Law enforcement and surveillance using cameras has overwhelmed human operators with gigabytes of video and still pictures. Searching manually for a certain license plate in a video is time consuming and error prone for current human operators. Having a human in the loop compromises the security and integrity of the system. Human beings are susceptible to intentional or non-intentional data tampering. An automatic system that detects and recognizes license plates is a must in areas like surveillance, identification of vehicles through video footage, automatic traffic violation systems and access control. Automatic License Plate Recognition (ALPR) is usually divided into two major parts, localization and recognition. Localization locates the part of the image which contains the license plate, while recognition identifies the individual characters on the assumption that the image being identified is that of a license plate.

In this chapter, we present three algorithms to recognize characters in Saudi license plates. All algorithms rely on processing information from lines strategically drawn through a character. The first algorithm calculates the number of peaks for each line. A peak is a place in the line where the pixels’ color changes from black to white. This algorithm is based on an earlier version (Obeid & Zantout, 2007; Obeid et. al, 2007) used to recognize characters in Lebanese plates. The second algorithm calculates the pixels density for a specific crossing line in a character. Pixel density is defined as the number of pixels having a specific intensity level to the total number of pixels in a line. The third algorithm calculates the position of the peaks introduced in the first algorithm rather than only their numbers. For each algorithm, the features were calculated for all possible letters and numbers. A study was then done on the best way to be able to differentiate between the letters and numbers based on the feature of the algorithm. Some of the methods had difficulty differentiating between a pair (or more) of characters. In general, methods that have high processing requirements were able to distinguish between characters while methods that were computationally simple had some difficulty with some pairs.

In this chapter, two simple but effective algorithms are also presented for locating plates in an image that contains Saudi license plates. The algorithms are adaptations of the algorithm presented in (Obeid et. al, 2007) for Lebanese license plates. In the first method, the image is segmented into labeled objects. Then the adjacency relationship between the different objects is studied to identify the part of the image which contains a license plate. The algorithm does not require recognizing the objects before locating the license plate. In the second method, instead of relying first on the alignment, the segmented objects are identified first by applying the three different recognition algorithms (Almustafa et.al, 2010; Zantout & Almustafa, 2011) to each object deemed possible to be a license plate component. A lot of ambiguity existed in the characters even after applying the recognition algorithms. This prompted the use of vertical alignment between objects to disambiguate between possibilities of characters. Although this method uses alignment between objects, however it only checks for vertical alignment between pairs of objects rather than horizontal alignment between up to six objects. This enables the method to be both simple and accurate. The major drawback of the method is that it is applicable only to new Saudi license plates. This was not considered a serious drawback since old Saudi license plates are due to be phased out totally by the end of year 2011.

In the next section, literature is reviewed for similar work done for license plates in general. Then, a description of Saudi license plates is given for readers unfamiliar with this kind of license plates. This information is imperative for understanding why certain decisions were made and thresholds were chosen. In the following section, three algorithms to recognize characters in Saudi license plates are presented along with results of applying these methods to actual Saudi license plates. Various thresholds used are presented along with a discussion of the results of applying the methods to actual license plates. Next, algorithms for plate localization are presented. The chapter concludes with a summary of achievements and suggestions for future research.