基于视频模式识别的快速超分辨率重建

Fast Super-resolution Reconstruction for Video-based Pattern Recognition

针对传统超分辨率重建方法运算量过大的问题,提出一种快速超分辨率重建算法。该算法在快速最大后验概率(MAP)重建算法的基础上,针对车牌识别的特定应用进一步简化了代价函数。该算法从低分辨率的图像序列中提取出高分辨率的图像。这些高分辨率的图像不仅极大地提高了车牌的识别率,也降低了获取高分辨率图像所需成像系统硬件的要求。移动车辆视频的盲重建实验表明,该算法能够有效地提高车牌图像的质量,而其运算量要远远小于其他的超分辨率重建方法。较低的运算量使得该算法能够有效应用于实时系统。

To reduce the computational cost of super - resolution reconstruction, a fast super - resolution reconstruction al- gorithm is proposed. For the special application of License Plate Recognition,it uses a further simplified cost function which is used in the fast MAP - based （Maximum a Posteriori） algorithm. The algorithm abstracts the resolution of reconstruction image from low resolution frame sequences. The result photos can greatly promote the recognition rate, and reduce the hard- ware requirements for imaging system producing higher resolution images. Blind reconstruction experiments using real videos of moving vehicles show that the proposed algorithm can efficiently enhance the quality of plate images, while its computational cost is quite lower compared with existing super - resolution methods. With low computational cost, this reconstruction method is suitable for real - time identification system.