非抽样复Contourlet变换的构造及其图像去噪应用

Construction of nonsubsampled complex Contourlet transform and its application to image denoising

为克服Contourlet变换由于缺乏平移不变性而在图像去噪等应用中存在的局限性,并利用其高度的方向性和各向异性,对原始的Contourlet变换加以改进,构造一种非抽样复Contourlet变换.该变换利用二维双树复小波变换和非抽样方向滤波器组分别进行多分辨率分析和方向分解,从而实现了Contourlet的复数变换.对图像去噪的实验结果表明,该变换除具有低冗余度和平移不变性外,还具有更丰富的方向分量,能够在去噪过程中有效地抑制伪Gibbs现象,更好地保护图像边缘和纹理等细节.其PSNR值和视觉质量均优于一般的去噪方法.

A nonsubsampled complex Contourlet transform was constructed to conquer the limitation of normal Contourlet transform, and its directionality and anisotropy were used. The proposed transform used two dimensional dual-tree complex wavelet transform for multiresolution analysis and nonsubsampled directional filter banks for direction analysis to realize the plural transform. Experiment result shows that the nonsubsampled complex Contourlet transform has the characteristics of low-redundancy and translation invariance, as well as more abundant direction components, so it can restrain Gibbs-like artifieials around edges in the course of denoising, and preserve more image details and textures efficiently. Its performance in both PSNR value and visual quality exceeds existing techniques.