Surfacelet联合多向Cycle Spinning的图像去噪方法

An image denoising method combining surfacelet transform and multidirectional cycle spinning

为了有效地保留去噪后图像的细节和纹理,综合利用Surfacelet变换和Cycle Spinning,提出一种新的图像去噪方法.Surfacelet具有良好的多维信号处理能力,但是Surfacelet不适宜直接处理二维图像,该文把多幅加噪图像分别进行多个方向的Cycle Spinning并压成图像序列,生成后的多个序列合成一个图像序列,进行Surfacelet变换后,对其系数硬阈值去噪.实验结果显示,去噪后图像无Wavelet的伪吉布斯现象以及Contourlet的划痕效果,该方法能明显改善图像视觉效果,显著提高图像的PSNR值.

In order to effectively retain details and texture while de-noising images, a new method for image denoising was developed which brings the strong points of surfacelet transform and multidirectional cycle spinning. Surfacelet transform is good at multidimensional signal processing, but it does not directly adapt to processing 2-D images. Many noisy images were processed with multi-directional cycle spinning and compressed to image sequences. These images were synthesized into an image sequence, and then surfacelet transform was conducted to denoise the hard threshold coefficient. Experimental results indicated that this method of de-noising images avoids the pseudo-Gibbs phenomenon as well as wavelet and nick effects. The method yields obviously better visual effects and high peak signal-to-noise ratio （PSNR） values.