基于定向双边全变差正则化的文档图像超分辨率算法研究

针对文档图像超分辨率重建问题,在传统双边全变差(Bilateral Total Variation,BTV)正则化超分辨率算法的基础上,提出了一种基于改进BTV的文档图像超分辨率算法。该算法引入一个新的正则项,即笔画宽度的方向,并根据字符笔画的局部宽度和局部方向自适应地进行平滑处理；然后通过分析输入的低分辨率图像及其插值,使输出图像的局部笔画宽度接近于局部的笔画方向。这种信息被压缩到基于笔画宽度的方向全变分正则项中。通过最小化正则项和数据保真项的线性组合,重建了高分辨率的图像。与相关的文档图像超分辨率方法相比,所提方法在视觉图像质量和字符识别精度方面得到了显著的改善。

Research of Document Image Super Resolution Algorithm Based on Directional Bilateral Total Variation Regularization

For super resolution image reconstruction problem in the traditional document,based on bilateral total variation regularization super-resolution algorithm,a document image super resolution algorithm based on improved BTV was proposed.By introducing a new regularization term,which is the direction of the stroke width,the algorithm is adaptive to the local width and the local direction of the character strokes.Then,the proposed algorithm can make the local stroke width of the output image close to the local stroke direction by analyzing the low resolution image and its interpolation.This information is compressed into the directional total variation regularization term based on the width of stroke.By linear combination of minimization regularization and data fidelity term,the high-resolution images are reconstructed.Compared with the related document image super resolution method,the proposed method has been improved in the visual image quality and character recognition accuracy.