提出一种基于四树复小波包变换(QCWPT)复数高斯尺度混合(CGSM)模型的图像去噪新方法。含噪图像经多尺度QCWPT被分解成一个低频复数逼近子图和若干高频复数方向细节子图。在大尺度下,可认为低频逼近子图为信号复系数予以保留;对各尺度复数方向子图按照对数熵代价函数确定最优复小波包基。把能很好刻画小波系数边缘分布形状和局部邻域强相关性的高斯尺度混合(GSM)模型扩展到复小波域,新的CGSM模型具有很好刻画图像幅值和相位信息的能力,并利用该模型对复系数进行贝叶斯最小均方(BLS)估计,从而实现去噪目的。实验结果表明,无论是峰值信噪比(PSNR)指标还是视觉效果,本文方法的去噪性能均好于双树复小波变换(DCWT)、QCWPT和小波域GSM模型去噪,并且在有效去噪的同时,具有很好的图像边缘和细节保护能力。
A new image noise suppression method using complex Gaussian scale mixtures(CGSM)in quad-tree complex wavelet packets transform(QCWPT)domain is presented.The noisy image is decomposed into a low frequency approach sub-image and some high frequency directional sub-images via the QCWPT.In the large sacle,the low frequency approach complex sub-image is looked as signal coefficients and retained unchangeably.The best complex wavelet packet basis is determined by using logarithm entropy cost function in high freq...