VEHICLE IDENTIFICATION SYSTEM BASED ON SINGULAR VALUE DECOMPOSITION

Abstract

Document Images are important in everyday lives. Document Image Recognition is required for information retrieving, searching, editing, and reporting of image text. There are several kinds of document images such as text, image, figure, signature, Certificate, type of ID card, car license plate, chassis number and so on. This thesis focuses to develop the Vehicle Identification System by using Vehicle Identification Number (VIN) images. The proposed system will provide to extract the detailed information to support decision making system for the Myanmar Customs Department. There have no two vehicles in real world have the same VIN. So, it can be used to identify the specific automobile as the unit ID like the fingerprint in human being. In this thesis, we propose Vehicle Identification System based on Singular Value Decoposition. This system includes five major steps: preprocessing, segmentation, or text area (VIN area) extraction, character segmentation, feature extraction and recognition. In the preprocessing step, Median filter is used to remove noises that can be in VIN image acquisition step. Average filter is also used to solve problems relating with contrast situation. As a second step, Canny edge detection algorithm and morphological processing are applied for possible VIN area extraction. Physical features such as size, shape, and ratio of width and height are considered for choosing final VIN (text) area extraction. In the next step, each character in the VIN area is segmented by histogram method. In the last step, for all the segmented characters, Singular Value Decomposition (SVD) features are calculated to recognize each character as recognition process. Finally, by combining the recognized characters, this work can easily give required information of a vehicle which belongs to detected VIN. Microsoft Access Database system is also used for VIN information system which can be used on ground because of working on real data. The system is implemented by C# and the result shows that the recognition method is feasible and easy to use in real world, so it can be put into practice. It can be easily extended to the real time vehicle examination and identification system by using mobile device thought over the network.