基于核范数的JPEG解码算法

A JPEG Decompression Algorithm Base on Nuclear Norm

针对低比特JPEG图像因量化过程中产生的量化噪声问题,提出一种核范数JPEG解码算法.首先基于自然图像的低秩性得到一个带无穷范数约束问题的低秩矩阵恢复模型,其次将约束凸优化问题转换为无约束优化问题,降低其计算难度.最后,利用经典的原对偶算法结合块匹配方法处理低秩矩阵模型,得到后处理JPEG解码图像.实验结果表明,该文算法比基于总变分后处理方法在去除量化噪声方面具有优越性.

Focused on the quantization noise in low bit JPEG images which is generated because of rounding the in the quantization process, a nuclear norm based JPEG decompression algorithm is proposed. Firstly, the model of low rank matrix approximation with an infinite norm constraint is constructed based on low rank property of nature images; Secondly, constrained convex optimization problem is transformed into a unconstrained optimization problem to reduce its calculation difficulty; Finally, the classical primal dual algorithm and blocking match method are exploited to solve the low rank matrix model so that a post processing JPEG images can be obtained. The experimental results demonstrate that the proposed method has the superiority of removing the quantization noise than total variation methods.