基于Curvelet多方向差和多尺度积的图像去噪

Image denoising based on multi-directional difference and multiscale products of curvelet transform

为滤除自然图像中的高斯白噪声并保持边缘、纹理信号,引入了Curvelet变换系数在方向间、尺度间和尺度内的相关性,提出了一种基于Curvelet系数相关性的多方向差和多尺度积图像去噪方法.首先根据图像纹理在Curvelet变换各尺度各子带上的方向信息,构造在方向倍增相邻尺度上的方向差,捕获方向间相关性,然后利用尺度间系数的多尺度积体现尺度间相关性,同时对各子带内系数采用局部聚集性反映尺度内相关性,最后综合这些相关性在图像真实信号和噪声上的不同表现来区分信号和噪声.实验结果表明:该方法具有更优的边缘保持和视觉平滑效果,与Curvelet阈值收缩方法相比,不但峰值信噪比有所提高,而且也较好地抑制了划痕现象.

An multi-directional difference and multi-scale products based on correlation method for im- age denoising was proposed to smooth out the Gaussian white noise and preserve edges and texture in image, using the dependencies among direction, inter-scale and intra-seale in the curvelet domain. First, multi-directional differences were constructed to capture the direction correlation on the adja- cent scales whose directions were double-increased, according to the direction information of image texture. Then, multi-scale products between scales and local aggregation in each sub-band were re- spectively utilized to represent the inter-scale and intra-scale correlations. Finally, true signals and noise were distinguished on the basis of these correlations. Experimental results indicate the new method has advantages in edge preservation and smooth effects. Compared with the curvelet shrinkage algorithm, this method can not only improve the peak signal to noise ratio （PSNR） but also suppress the scratches phenomena in the image denoising.