Malaysia’s Car Plate Recognition System

Phor, Joon Huang (2016) Malaysia’s Car Plate Recognition System. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Phor Joon Huang_FULL TEXT.pdf Restricted to Registered users only Download (1MB)

Abstract

Purpose: In partial fulfilment of the requirement for Bachelor of Science in Software Engineering, research on Characters Segmentation is carry out. The main objective is to isolate the image into several useful parts for further analysis purpose. Example: all of the car plate’s characters are segmented into several individual character instead of a bunch of characters which formed the car plate like from “WNW 2001” into “W”, “N”, “W”, “2”, “0”, “0”, “1”. Scope: Instead of whole image with car object, segmentation process will only perform on the image with car plate only. Only Malaysia’s car plate is accepted as an input to the segmentation process in this project. Both standard and non-standard car plate with black colour as background and characters in white colour is acceptable. Besides that, sufficient light, low background contrast environment must be fulfil and gap between the characters must not be too narrow. Once input is received, image pre-processing is performed right before segmentation process carry out. Methodology: Mixed research method is conducted in this project. A further explanation on mixed research method where mixed research method is also called multi-methodology that include unification of quantitative and qualitative approaches. Sequential exploratory strategy, one of the mixed research strategy is used whereby the qualitative research is carried out at the earlier phases of project and quantitative research will be at the latter phase of project. This Project was carried out using the image processing library, EMGUCV, Microsoft Visual Studio 2015, C# programming language, 32-bit Microsoft Windows 7 and above and Local storage of computer with 4GB of available hard-disk space. Development phases: For the initial phase, input is delivered from the previous phase and used to carry out the segmentation works. After that, image pre-processing is performed to eliminate the noises and convert the image to produce a binary image which only has two colour, black and white. The binary image was used by the selected algorithm, combined version of Connected-Components Analysis and Blob Detection. Blob Detection was first used to identify the characters’ region boundary and produce the coordination of each results. Based on the coordination value, Connected-Components Analysis capable to identify the location of the character and looking for the connected pixels and label the components. Once done, the components was draw on a bitmap object and act as the outputs of segmentation works. In the end, each isolated bitmap object is converted to an image object. Assessment criteria used: There are two assessment criteria, best suited threshold value and outputs accuracy being done on segmentation works. The first assessment was to determine the threshold value that can identify the highest number of characters in each car plate image which include the proposed algorithm and compare with using the other algorithm alone. While the second assessment was to evaluate the output’s accuracy based on the ground truth label produced manually on the car plate image and compared with the segmentation’s outputs that provided by the proposed algorithm and other existing algorithm. Results : Based on the performed assessment on segmentation works, the proposed algorithm achieved higher than other two existing algorithm no matter in the first or second assessment. For the strength, the proposed algorithm capable to avoid segmented the unnecessary or unintended objects or areas as the outputs. The main reasons for the proposed algorithm to achieve higher accuracy compare to other algorithm was because of its strength mentioned above. For the weakness, the threshold value defined has to be fixed which limit the coverage of the segmentation works. Conclusions and Recommendations : From the research perspective, current segmentation work consists of several limitation. Example: fixed threshold value, required sufficient light condition and hard to determine the car plate’s characters precisely if all types of Malaysia’s car plate are going to cover due to different in font style, size, and pattern. Future improvement are having on segmentation work are making threshold value dynamic instead of fixing on certain threshold value to process the car plate, capable to recover lost features on the car plate’s characters in order to extract the characters in complete shape and so on