一种新型轮廓波变换及基追踪降噪

An Improved Contourlet Transform and Basis Pursuit Image Denosing

由于轮廓波变换(Contourlet Transform)基图像存在不具备局部频率特性的缺陷,通过分析频域楔形支撑区域外部出现潜在混叠的原因,结合易操纵金字塔结构的优良特性,提出了一种新型轮廓波变换来克服传统轮廓波变换的缺陷.这种新型轮廓波变换采用易操纵金字塔替换原变换的拉普拉斯金字塔,保证了其平移不变性.通过非线性逼近实验和基图像分析,得出新型轮廓波变换能稀疏表示图像和避免频域混叠,建立的基追踪图像降噪模型结果表明新型轮廓波变换可以同时提高降噪图像的峰值信噪比和图像质量.

In view of contourlet transform basis images are not localized in the frequency domain~ After analyz- ing the cause of frequency domain aliasing and many fine properties of the steerable pyramid, we propose improved contourlet transform to overcome the shortage of original contourlet transform. The proposed transform employs the steerable pyramid instead of the laplacian pyramid to complete multiscale decomposition, which makes the trans- form has shift invariance. By doing nonlinear approximation experiment and analyzing basis images, We conclude that this transform can sparsely express images and avoid frequency domain aliasing. Building basis pursuit image denosing model results show that the proposed transform outperforms the original contourlet transform both in terms of peak signal noise rate （PSNR） and visual quality.