基于重加权低秩近似和高斯混合模型的图像块效应去除算法

针对离散余弦变换(Block-based Discrete Cosine Transform, BDCT)在图像解码器处产生的块重构伪影现象，提出了一种联合两种先验知识的图像去块效应算法，这两种先验知识分别是重加权低秩近似和高斯混合模型。该算法首先利用重加权低秩近似来增强图像块之间的局部结构和非局部的自相似性，有效地保留原始图像中更多的精细结构。其次，还利用高斯混合模型对块状伪影进行建模获得更可靠、更鲁棒的结果。通过在标准测试图像上的实验表明，提出的算法在主观视觉效果和客观评估方面均优于其他的块效应去除方法。

Image Deblocking via Reweighted Low-Rank Approximation and Gaussian Mixture Model

To remove the block reconstruction artifacts from image decoder, which are generated by block-based discrete cosine transform (BDCT), this paper proposes a block removing algorithm with two kinds of joint priors that are reweighted low-rank approximation and Gaussian mixture model respectively. Firstly, it adopts reweighted low-rank approximation to enforce the intrinsic local structure and nonlocal self-similarity existed among blocks from the image, which is believed to better preserve the fine structure existed in the original image. Then, the Gaussian mixture model is explicitly incorporated to model the blocking artifacts, which helps to obtain a more reliable and robust solution. Experimental results demonstrate that the proposed algorithm outperforms other comparison deblocking methods in both visual perception and objective evaluators.