快速去块效应的线性规划方法

Fast image de-blocking by linear programming

许多现有的图像压缩算法在高压缩比下会产生恼人的块效应,消除块效应的后处理方法一直以来都是图像处理领域的重要研究方向。消除块效应可以认为是从不准确的采样数据出发,尽可能恢复原始图像,这也是压缩传感理论所做的。因此利用压缩传感理论,给出了一种新的去块效应方法,将去块效应问题归结为一个无需调校任何其他参数的线性规划问题,最终采用GPU实现,得以快速求解。大量的实验结果表明,该方法能快速有效地去除块效应,改善了图像的视觉效果,同时提高了图像的PSNR。

Compressed images may have block artifacts at low bit rates in many image compression algorithms. Postprocessing methods for image de-blocking are the most practical solution for removing block artifacts, since this does not require any changes to the existing standard codees. Image de-blocking can be considered as recovering the ground-truth image from inaccurate samples. It is exactly what compressive sensing does. According to this, we take advantage of compressive sensing theory to remove block artifacts. As a result, we convert the image de-blocking problem to a linear programming problem in which no parameters are required to be tuned. Finally, our approach can be performed fast using a GPU implementation. Our experiments show our approach can effectively remove block artifacts from compressed images, improving the visual quality and PSNR.