基于流形学习的车牌图像超分辨率算法

为了解决视频监控系统中车牌图像分辨率较低、车牌字符难以辨识的问题,提出一种基于流形学习的车牌图像超分辨率重建算法。首先学习训练样本库中高、低分辨率图像之间的映射关系,然后利用线性判别分析(Linear Discriminant Analysis,LDA)算法提取图像特征,并利用流形学习中的局部线性嵌入(Locally Linear Embedding,LLE)算法对特征向量进行参数建模,最后通过特征映射关系获得高分辨率图像。实验表明,该方法对监控系统中的低分辨率车牌图像具有较好的超分辨率复原效果,不仅提高了字符的可读性,而且具有更高的峰值信噪比。

A license plate super-resolution reconstruction algorithm based on manifold learning

In order to address the problem of low resolution in surveillance video which causes the difficulty in recognizing license plates,a new license plate image super-resolution reconstruction method based on manifold learning is presented in this paper.Firstly,the mapping between lowresolution images and high-resolution images in training set is obtained by learning method.Secondly image feature vectors are extracted by linear discriminant analysis（LDA）algorithm and its parameters are modeled by locally linear embedding（LLE）algorithm.Finally the high resolution image is reconstructed by the mapping relation.The experimental results show that the proposed algorithm has good reconstructed performance for real surveillance system,which enhances video quality in terms of peak signal to noise ratio（PSNR）and improves interpretation of license plates.