在利用曲波对图像进行分解的基础上,提出了一种考虑不同尺度分量的曲波系数间是相互关联的系数概率密度函数(PDF)模型,不同尺度的系数选择依赖其父层系数的大小,从而保证在去除噪声的同时,保留图像的纹理边缘等细节信息,然后对图像进行重构。实验表明,该方法能有效地去除噪声,同时较好地保留了图像的纹理和边缘等细节,而且对图像质量有进一步的改善。
Based on the relativity of the different scale coefficients,we proposed a model to model the curvelet coefficients for denoising. The different scale coefficients depend on the previous coefficients,so as to eliminate the noise in the images and keep the image details. The experiments show that the method can remove the noise efficiently and keep the details very well,and it can make a great improvement of image quality.