基于小波系数多尺度随机过程模型的去噪方法

Wavelet image denoising based on multiscale stochastic process coefficient model

用多尺度随机过程对小波图像系数进行建模,并在此基础上提出了基于多尺度随机过程模型的小波图像去噪方法。通过阈值判断和邻域判断相结合的方法区分出对应边缘处的系数。对边缘区小波系数树估计多尺度随机过程的参数,利用多尺度滤波器对小波系数进行估计,对非边缘区的小波系数则采用阈值萎缩方法进行估计。该方法很好地刻画了边缘区小波系数跨尺度的行为,可以很好地保持图像边缘;而且还给出了估计误差的方差,利于理论分析。实验表明:该方法的去噪误差要优于Sureshrink法,而且对图像边缘的保护更利于后续的图像分割和轮廓跟踪。

The multiscale stochastic process was used to model wavelet coefficients and to derive a new wavelet image denoising method. Thresholding decision making was combined with a neighbor decision rule to classify the wavelet coefficients into two classes, coefficients in sharp regions such as edges and coefficients in smooth regions. Then, a multiscale filter was used to estimate the noiseless coefficients in the first class, while those in the second class were estimated with wavelet shrinkage. The new model accurately describes the crossscale features of the coefficients in the first class, which preserves the edges. The error variance was also derived for further analyses. Experiments showed that the new method not only provides less reconstruction error than Sureshrink, but also facilitates further image processing such as image segmentation and contour tracing due to the good edge preservation.