Abstract: An improved image denoising method of first optimization and last classification in the wavelet domain is proposed.It is an improvement of the existing denoising method NeighShrink.The proposed method determines an optimal threshold and the window of the neighborhood using stein unbiased risk estimation in the wavelet domain for each sub-band.According to neighborhood threshold size，it divides wavelet coefficients of sub-band into“small”coefficients or“big”coefficients.Those“small”coefficients are set to zero，whereas those “big” coefficients are modeled as zero-mean Ganssian random variables with high local correlation and the estimation of the true coefficients are obtained by minimum mean squared error criterion.The experimental results show that the proposed method obviously outperforms the NeighShrink method in the peak signal to noise ratio.At the same time，it effectively preserves image texture information，and has better visual effects.

Key words: first optimization, last classification, image denoising, minimum mean squared error

摘要： 提出一种先优化后分类改进的小波域图像去噪方法。该方法是对现存NeighShrink去噪方法的改进，用stein的无偏风险估计，在小波域每一个子带确定一个最优的阈值和邻域窗口；根据邻域阈值的大小，将子带内的每个小波系数划分为“小”系数或“大”系数；对“小”系数直接置零，对“大”系数采用一种具有局部空间强相关性的零均值高斯模型，通过最小均方误差准则得到真实系数的估计。实验结果表明，该方法在峰值信噪比指标上明显优于NeighShrink方法，同时有效地保存了图像的纹理信息，视觉效果较好。

关键词: 先优化, 后分类, 图像去噪, 最小均方误差