基于快速离散曲波变换的图像去噪算法

Image denoising based on fast discrete curvelet transform

Curvelet变换可以更好地表示曲线奇异函数的异向性及图像边缘,因此更适合于多尺度图像去噪。针对传统阈值法存在的不足,在分析wrapping方法的快速离散曲波变换基础上,提出结合Cycle Spinning循环平移方法的菱形块阈值规则去噪法,并自适应地对不同的Curvelet子块进行阈值化。该方法可以消除由于Curvelet变换缺乏平移不变性而产生的图像失真,并且更好地利用曲波系数的相关性。实验结果表明,该方法与传统的小波去噪、曲波硬阈值去噪、曲波软阈值去噪、曲波软硬阈值折中法去噪相比,使得去噪图像的峰值信噪比更高,视觉效果更好。

Curvelet can reflect anisotropy of singular function and image edges; therefore it is better suitable for multiseale image denoising. According to the defects of the traditional thresholding methods, the diamond-shaped pieces thresholding algorithm combined with cycle spinning algorithm was proposed after analyzing the fast discrete eurvelet transform based on wrapping algorithm, and the curvelet transform coefficients in different subbands were filtered with adaptive thresholds. The method carl avoid image distortion due to the lack of translation invarianee of eurvelet transform, and earl make use Of correlation of eurvelet coefficients better. Experimental results show that the proposed method.yields denoised images with higher PSNR and better visual effects compared with the traditional wavelet denoising algorithm, the hard-threshold denoising method, the soft-threshold denoising method and the method between soft and hard thresholding based on curvelet transform.