基于多字典学习的车牌图像超分辨率重建方法

摘要

车牌识别作为智能交通系统中的关键技术，具有十分重要的应用前景。本文提出了一种改进的车牌识别算法。采用K-均值算法对高分辨率车牌图像样本进行聚类，并按类学习相应的子字典，从而有效地减少了字典中原子的数目，提高了算法的运行速度。再采用自适应的正则化参数法和反向投影法对图像进行全局优化，既保留了图像的更多细节特征，又起到较好的去噪效果。仿真实验结果表明，该算法从主观和客观效果上均有效地改善了重建的车牌图像质量。%In the process of image super resolution reconstruction, use the same complete dictionary of faults is not considering local characteristics and large computation. An improved reconstruction algorithm of license plate image is proposed to solve the problem. Using adaptive regularization parameter based on the reconstruction algorithm of multi-dictionaries learning, not only reserve the more characteristics of image and get rid of the noise better. Finally, global optimization using reverse projection, make the result more natural. Through experimental evaluations, the algorithm effectively improves the reconstruction of the license plate image quality.