查询结果:   张奇,吕晓琪,李银辉,于荷峰,候贺.L1和L2混合范式超分辨率重建的车牌识别[J].计算机应用与软件,2016,33(7):176 - 180.
中文标题
L1和L2混合范式超分辨率重建的车牌识别
发表栏目
人工智能与识别
摘要点击数
791
英文标题
LICENSE PLATE RECOGNITION SYSTEM WITH MIXED L1 AND L2 NORM SUPER-RESOLUTION RECONSTRUCTION
作 者
张奇 吕晓琪 李银辉 于荷峰 候贺 Zhang Qi Lü Xiaoqi Li Yinhui Yu Hefeng Hou He
作者单位
内蒙古科技大学信息工程学院 内蒙古 包头 014010     
英文单位
School of Information Engineering,Inner Mongolia University of Science and Technology,Baotou 014010,Inner Mongolia,China     
关键词
L1范式 L2范式 超分辨率重建 方向梯度直方图 支持向量机
Keywords
L1 norm L2 norm Super-resolution reconstruction Histogram of orientation gradient Support vector machine (SVM)
基金项目
国家自然科学基金项目(61179019);内蒙古科技大学创新基金项目(2014QDL045)
作者资料
张奇,硕士生,主研领域:图像处理。吕晓琪,教授。李银辉,硕士生。于荷峰,硕士生。候贺,硕士生。任国印,讲师。 。
文章摘要
针对视频中低分辨率图像的车牌识别准确率低的问题,提出一种结合L1和L2混合范式的序列图像超分辨率重建的车牌识别技术。首先对序列低分辨率图像进行L1和L2混合范式超分辨率重建,其次对重建后得到的一帧高分辨率图像进行基于HSV颜色模型车牌定位,然后对分割出的字符采用方向梯度直方图和支持向量机相结合的方法进行车牌识别。实验结果显示提出的算法对车牌中的字符识别效率高达96%,对比于传统的基于特征匹配和BP神经网络的车牌识别算法对字符的识别有明显的改善。结果表明,通过L1和L2混合范式的超分辨率重建处理,将方向梯度直方图和支持向量机相结合的识别方法对车牌中的字符有较好的识别效果。
Abstract
Focusing on the problem of low recognition accuracy of license plate in low resolution video images, we put forward a license plate recognition technology which is based on sequence image super-resolution reconstruction with mixed norm combining L1 and L2. First, we made super-resolution reconstruction using mixed L1 and L2 norm on sequence low resolution image. Secondly, on a high-resolution image derived from the reconstruction we located the license plate based on HSV colour model, and then identified the license plate by applying the method of combining orientation gradient histogram and support vector machine on the segmented characters. Experimental results showed that the recognition efficiency of the proposed algorithm on the characters in licence plates was as high as 96%, and had a significant improvement in character recognition compared with the traditional feature matching-based algorithm and BP neural network licence plate recognition algorithm. The result demonstrated that through the processing of mixed L1 and L2 norm super-resolution reconstruction, the recognition method combining the orientation gradient histogram and support vector machine has better recognition effect on license plate characters.
下载PDF全文