A Practical Plate Character Recognition Algorithm Study Based on Thinning Characters

A practical thinning-based license plate recognition method is presented in this paper. This method combines the advantages of the projection, structure needle and symmetry features to recognize characters. First it thins the binary character image with the method based on the index table, and gets the peak feature which includes value and location from both the horizontal and vertical projections of the thinned image. The initial recognition could be achieved with these features. For some characters, their projections are too similar to be distinguished. These require the needle features to be extracted to get further recognition. Finally, the local symmetry features are extracted recognizing the similar characters. In this paper the test of 1732 characters (numbers and English letters) taken in various illumination conditions of license images resulted in the correct recognition rate over 95%.