This paper presents a novel approach to locate multiple motorcycles license plates with different orientations and illumination. In the first stage, some pre-processing steps through visual image processing are used to remove the unwanted regions and to intensify the regions of interesting. In the second stage, we use the properties of high density which license plate has on the dilation map to verify. In the third stage, in order to locate the license plates, we utilize some geometric rules and texture information to identify the candidate regions. Finally the license plates will be labeled through Connected Component Analysis. The consuming time of proposed algorithm is about 0.83 seconds. Experiment shows that superior result is achieved comparison to other method.
第18屆電腦視覺圖學暨影像處理研討會論文集=Proceedings of the 18th IPPR Conference on Computer Vision, Grapfics and Image，5頁