A novel accurately automatic license plate localization method

Automatic license plate (LP) recognition is one of the most promising aspects of applying computer vision techniques towards intelligent transportation system. In order to recognize a license plate, the localization of it must be detected first. Obviously, accurately localize the plate from a vehicle image, is a crucial step of an ALPR system, which greatly affects the overall recognition performance. In this paper, we propose a color edge based algorithm in the preprocess step which not only consider the gray level discontinuity but also some color information between the LP region and non-LP region. Moreover, we adopt a morphological technique to extract edges instead of Sobel or Marr operator which is used mostly in other works. We also propose a concentric double window filtering method (CDWF) to handle the raw edge map. Then we choose the candidate LP region based on edge density. After the rough localization, we getting the LP´s exact region by means of profile projections, and then using a mask map to represent the region. The segmentation result which is sent forward to LP recognition stage will improve further processing´s efficiency. Considering some noise may be exist in the profile projection, the raw projection map may be inaccurate, fragmented or incomplete in some position, we could apply morphological reconstruction to the mask map. Finally we demonstrate our whole technique´s performance with experiments.