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

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.