The focus of this thesis is on reliable and robust license plate recognition (LPR). The technology is currently in operation that both the quantity and quality-based approaches are needed. The entire procedure of license plate recognition consists of six steps: (1) image acquisition, (2) image pre-processing, (3) plate locating, (4) character segmentation, (5) character recognition, (6) output. However, when a road is uneven and slant, a vehicle will be shaky while running; consequently, the plate is also unstable and tilted with an angle of rotation at this moment. Meanwhile, a surveillance camera is very difficult to capture an effective image so that the plate is hard to be located and recognized. Due to this existing problem, the contributions of this thesis are: (1) plate tilt correction using Hough transform; (2) GNN-based plate number recognition. The novelties of this thesis are to improve the robustness and reliability of license plate recognition through literature review as well as evaluate those existing technologies.