基于连续淘汰的快速自相似错误隐藏算法

Fast best neighborhood matching algorithm based on successive elimination

为了减少自相似错误隐藏算法中搜索过程的计算复杂度,提出了一种基于连续淘汰的快速自相似错误隐藏算法。该方法在已有自相似错误隐藏算法的基础上,利用域块和值块相邻像素之间的关系,对域块的运动纹理特征进行分析以实现搜索过程中窗口尺寸的自动伸缩。利用连续淘汰方法,通过计算值块和域块各自像素和的差与像素差的和,得到自相似块搜索过程中域块的最佳匹配值块间需要具备的必要条件,通过比较直接淘汰掉搜索过程中不满足条件的值块。实验结果表明,该算法有效降低了自相似错误隐藏算法的计算复杂度,恢复后的图像质量优于各种插值算法,并有效消除了错误传递现象。

For decreasing the computational complexity of searching process in best neighborhood matching method, a fast error concealment algorithm using successive elimination was proposed. Based on the existing algorithm, the motion and texture information on the range block was obtained by analyzing the neighboring domain blocks, and the size of searching window was changed adaptively. The successive elimination method was extended to find the best matching domain block in the matching process. By computing the sum absolute difference （SAD） of range block pixels and domain block pixels, successive elimination compared this SAD value with the difference value between the sum of range block pixels and the sum of domain block pixels. The restriction was regarded as decision boundary for the best matching range block, and the redundant matching blocks were eliminated directly. Experimental results show that the fast algorithm reduces the computational cost efficiently compared with the conventional best neighborhood matching method. Better image quality is achieved than that of interpolation algorithms, and the error propagation is avoided.