基于低秩矩阵分解的批量扫描文档图像纠偏

扫描文档图像纠偏的关键是对图像偏转角度进行快速准确的估计。传统的基于图片自身纹理结构的算法，如Hough变换、Radon变换，不仅易受文档自身特殊结构或噪声影响，而且单幅图像纠偏的平均耗时较长。提出了一种基于低秩矩阵分解理论扫描文档图像的批量纠偏方法，该方法将批量图像构造成一个较大的矩阵，通过迭代对每一列进行适当地旋转，达到矩阵具有较低秩的目的，进而实现对每副图像偏转角度的恰当估计及纠偏。实验结果表明，该方法不仅具有较高纠偏的精度，而且单幅图片的平均耗时也小于现有的图片纠偏算法。

Skew correction of batch scanned document images based on low-rank matrix decomposition

The most important of skew correction for scanned document image is to estimate the skew angle. Traditional methods are mostly based on its texture structure check, such as Hough transformation, Radon transformation and so on. However, it not only is affected easily by its special texture structure or other noise, but also costs much more time for each image. In this paper, a skew correction of batch scanned document images method is proposed based on low-rank matrix decomposition, which to seek an affine rotation transformation that can be used to implement the correction. As experiment illustrated, the method not only has high precision of image rectification, and the average time is less than the existing algorithms of image correction.