一种用于车牌识别的图像超分辨率重建技术  

IMAGE SUPER-RESOLUTION RECONSTRUCTION FOR LICENCE PLATES RECOGNITION

作　　者：刘芳华[1] 倪浩[1] 阮若林[2] 王建峰[3] 

机构地区：[1]湖北科技学院电子与信息工程学院,湖北咸宁437100 [2]湖北科技学院生物医学工程学院,湖北咸宁437100 [3]湖北科技学院网络管理中心,湖北咸宁437100

出　　处：《计算机应用与软件》2016年第11期208-211,262,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61271256);湖北省高等学校优秀中青年科技创新团队计划项目(T201513);湖北省自然科学基金项目(2015CFB452);湖北省教育厅科研计划指导性项目(B2015080);湖北科技学院校级科研项目(KY13048)

摘　　要：获取的车牌图像因分辨率过低、过量模糊和噪声等原因会导致其图像质量较低,影响了车牌识别的准确率。为了提高车牌识别的准确率,采用基于学习的超分辨率重建算法增强低质车牌图像。引入在线字典学习方法训练超完备字典,并制作适合于车牌超分的训练图集,根据低质车牌图像重建高分辨率车牌,按照既定的模板匹配方法进行车牌识别。实验表明,超分方法的PSNR和SSIM比经典的SCSR(Sparse Coding Super-Resolution)法都有明显提升,车牌识别率也比SCSR提高了5.0%。可见,所提出的算法较好地增强了低质车牌的图像质量,有效地提高了识别率。Some captured license plate images are low-quality because of the low-resolution,blur and noise which affects the recognition accuracy. In order to prove the recognition accuracy,the proposed method employs the sample-based super-resolution to enhance the low-quality licence plate images. It produces the training image set which is suitable for plate recognition and introduces Online Dictionary Learning to get the over complete dictionaries. After the reconstruction of the high-resolution plate image from a low-quality one,the defined template matching method recognizes the plate numbers well. The experiments show that the PSNRs and SSIMs of the proposed method are better than the classical SCSR. The percentage of the plate recognition accuracy is 5. 0% higher than SCSR. So the proposed method can enhance the low-quality images and prove the recognition accuracy effectively.

关 键 词：超分辨率 车牌识别 在线字典学习 稀疏编码 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]