论文部分内容阅读
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 multi-scale stochastic process is used to model the wavelet image coefficients. On the basis of this, a wavelet image denoising method based on multi-scale stochastic process model is proposed. Through the combination of threshold judgment and neighborhood judgment, the coefficient at the corresponding edge is distinguished. The coefficients of multiscale stochastic processes are estimated for the wavelet coefficient tree in the edge region, the wavelet coefficients are estimated by using multi-scale filter, and the threshold shrinkage method is used to estimate the wavelet coefficients of non-edge regions. The proposed method well characterizes the cross-scale behavior of wavelet coefficients in the edge region, and can well preserve the edge of the image. Moreover, the variance of the estimation error is also given, which is good for theoretical analysis. Experimental results show that the proposed method is superior to Sureshrink algorithm in denoising error, and the image edge protection is more conducive to the subsequent image segmentation and contour tracking.