摘要
针对Contourlet变换存在的频谱混叠,提出了一种抗混叠塔型变换,即Non-aliasing Pyramidal Transform-NA-D变换.NAP变换由抗混叠塔式滤波器组(Non-aliasing Filter Banks:NPFB)和方向滤波器组(Directional Filter Banks:DFB)组成,NPFB首先将图像分解为多个不同分辨率的细节子带和一个低频子带,DFB再将各细节子带分解为方向子带.通过设计满足Nyquist采样定理的滤波器,NA-D变换有效地抑制了Contourlet变换的频谱混叠,基函数不仅具有"多分辨率"、"多方向"、"局域性"等特性,满足各向异性尺度关系,而且空频域正则形和局域性均明显优于Contourlet变换.硬阈值去噪实验的结果表明,NAP变换能够更为稀疏的表示图像,并在去噪性能上较Contourlet变换有较大提高.
A non-aliasing Pyramidal transform, namely NAP transform,is presented to avoid the frequency aliasing of Contourlet transform. NAP transform consists of non-alia,sing pyramidal filter banks （NPFB） and DFB. In this paper, we firstly represent the definition and specific parameter setting of NPFB in the frequency domain. NPFB decomposes image into an approximation subband and several detail subbands with different resolutions; whilst DFB decomposes the detail subbands into directional subbands. Not only has the basis function of NAP transform a variety of characteristics such as multi-resolution,multi-direction,localization to meet demands of anisotropy scale relations, but also are their regularity and localization better than those of Contourlet transform in the spatial domain. Results from the experiments of image denoising show that NAP transform can represent image more sparsely and has a significant improvement from the aspect of image denoising, compared with Comourlet transform.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2009年第11期2510-2514,共5页
Acta Electronica Sinica
基金
重庆市自然科学基金(No.2009BB2188)
