Simplified Local Binary Pattern Descriptor for Character Recognition of Vehicle License Plate

Local Binary Pattern (LBP) is a powerful texture descriptor for its tolerance against illumination changes and its computational simplicity. The basic LBP encodes 256 feature patterns in a 3×3 neighborhood, but not all the patterns are effective for classification. In this paper, we propose a simplified LBP(S-LBP) which produces optimal patterns by using the best coding principle for classification. Meanwhile, we combine S-LBP and Mahalonobis distance in solving the practical problem of character recognition in Chinese license plate. Experimental results demonstrate the effectiveness of our method for vehicle license recognition comparing with other popular methods.