Abstract: This paper presents a novel algorithm for license plates location. The input image is segmented by block projection to obtain potential regions with plate characters. It picks up candidate regions of plates by clustering those potential regions with semantic restrictions. False plate regions are deleted according to texture characteristic. 3000 images are processed with an accuracy of 97.1%. At the same time, the average time is less than 35 ms. The results prove that the algorithm is effective enough to support real-time license plate recognition.

Key words: license plate location, block projection, semantic restrictions, real-time

摘要： 以提高车牌定位实时性和同时搜索多个车牌为目的，研究并提出一种基于分块投影和语义约束的车牌快速定位算法，通过投影分割处理得到候选车牌字符块，根据语义约束对字符块进行聚类得到候选车牌区域，根据纹理特征过滤伪车牌。对3 000幅图像进行实验，定位成功率为97.1%，平均定位时间小于35 ms。实验证明这种算法适应性强、速度快，能够满足车牌实时识别的需求。

关键词: 车牌定位, 分块投影, 语义约束, 实时性