一种改进的非下采样轮廓波变换图像去噪算法

Improved Image Denoising Algorithm Based on Non-subsampled Contourlet Transform

优化图像去噪问题,在非下采样轮廓波变换图像去噪中,收缩阈值的确定仅依赖变换子带系数的幅值,使得过多图像系数和噪声系数一并去除,导致滤波图像模糊。从检测变换子带几何结构出发,引入自蛇模型对子带系数作几何结构检测并抑制噪声后,估计双阈值将子带系数划分为三类并作不同处理,实现对噪声系数的去除和对图像系数的保护。实验结果表明,相对现有典型算法,改进算法获得的峰值信噪比提高了0.1-0.9dB,图像系数被更好识别和保留,滤波图像中边缘与区域细节损失减少,提高去噪效果,保留图像的有效信息。

Many image denoising algorithms based on non -subsampled contourlet transform （NSCT） remove image coefficients excessively due to the decision of shrinkage thresholds solely depends on coefficient amplitudes, which leads to blurred edges in denoised images. This paper started with geometric structure detection of a NSCT subband, and the self - snake model was introduced to conduct the detection and noise coefficient suppression. And then, two thresholds were estimated to classify the coefficients in a subband into three categories, they would be processed differently for the purpose of image coefficients protection with removal of noise. Experiments results indicate that, compared to current typical algorithms, the proposed algorithm recognizes and protects weak image coefficients effectively. This capability produces higher PSNR values between 0. 1 to 0. 9dB and better protection of geometric structures, which leads to less blurred denoised images.