An Efficient Feature Extraction Approach for License Plate Recognition

抄録

It is a known fact that character recognition is a key to a high recognition rate for a license plate recognition. Hence, this paper presents a novel feature extraction approach in an attempt not merely to well maintain a high recognition rate but also to reduce the computational load considerably. As opposed to a conventional counterpart, feature extractions are performed on a column and on a row basis in this novel approach. With a total of M × N blocks over such divided character, the dimension of a feature vector is reduced from M × N in a conventional approach to M + N in this work. It is experimentally demonstrated that the recognition accuracy rate is well maintained with a 91% computational load reduction through k-means clustering algorithm in the case of M=36, N=16. In other words, the use of this algorithm enables a handheld device to meet the energy saving requirement for extended operations.