基于噪声估计的自适应块匹配和三维滤波降噪算法

发布时间：2018-07-08 17:53

本文选题：图像降噪 + 块匹配和三维滤波(BMD)算法　；参考：《光电子·激光》2017年06期

【摘要】：针对经典的块匹配和三维滤波(BM3D)降噪算法中最为核心的噪声水平(方差)参数在使用中需要人工手动设置极大影响了降噪效果并限制了它的应用,提出了一种新的基于自然场景统计(NSS)的噪声水平特征矢量和支持向量回归(SVR)技术的快速噪声水平估计算法并应用于经典BM3D算法中,使之转变为自适应降噪算法(Adaptive BM3D)。本文算法首先利用小波变换对图像进行不同尺度和不同方向的分解,提取各子带滤波系数并用通用高斯分布模型(GGD)建模,以模型参数构成反映噪声图像噪声水平的特征矢量;然后用SVR方法在大量噪声图像样本上进行训练获得图像噪声水平预测模型。实验表明:改进后的ABM3D算法实际图像降噪效果比BM3D算法获得进一步提升,并且仍然保持了非常高的执行效率,相对于当前各主流算法具有明显的竞争力。
[Abstract]:In the classical block matching and 3D filtering (BM3D) denoising algorithm, the manual setting of the noise level (variance) parameter, which is the core of the classical block matching and three dimensional filtering (BM3D) denoising algorithm, has greatly affected the noise reduction effect and limited its application. A new fast noise level estimation algorithm based on the feature vector and support vector regression (SVR) of natural scene statistics is proposed and applied to the classical BM3D algorithm to transform it into an adaptive noise reduction algorithm (Adaptive BM3D). Firstly, the wavelet transform is used to decompose the image in different scales and in different directions, and the filter coefficients of each subband are extracted and modeled with the general Gao Si distribution model, and the feature vectors reflecting the noise level of the noise image are constructed by the model parameters. Then SVR method is used to train a large number of noise image samples to obtain image noise level prediction model. The experimental results show that the actual image denoising effect of the improved ABM3D algorithm is further improved than that of the BM3D algorithm, and it still maintains a very high execution efficiency, which is obviously competitive compared with the current mainstream algorithms.
【作者单位】：南昌大学信息工程学院;
【基金】：国家自然科学基金(61662044,61163023,61379018) 国家级大学生双创项目(201510403030)资助项目
【分类号】：TN713;TP391.41

【相似文献】