Abstract:

In order to facilitate character segmentation and recognition in a license plate recognition system, the tilting license plates in the located images should be corrected in both the horizontal and vertical directions. A vertical
tilt correction method by minimizing variance of coordinates of the projection points is proposed. Coarse character segmentation is performed after horizontal correction and upper and lower border delineation. The character points are projected along the vertical direction after shear transform. Two closed-form expressions of the shear angle are derived based on minimizing variance of coordinates of the projection points. Thus the vertical tilt angle is directly determined and the license plate images corrected using the vertical tilt angle. Experimental results show that, compared to the vertical tilt correction methods of the license plate based on rotation projection and line fitting, the proposed method has high accuracy, requires less computation, and is robust. The proposed method has been applied in practical engineering system.

Key words: vehicle license plate recognition, vertical tilt correction, shear transform, vertical projection, minimum variance