双密度双树复小波变换的局域自适应图像去噪  

Local adaptive image denoising based on double-density dual-tree complex wavelet transform

作　　者：龚卫国[1] 刘晓营[1] 李伟红[1] 李建福[1] 

机构地区：[1]重庆大学光电技术及系统教育部重点实验室,重庆400030

出　　处：《光学精密工程》2009年第5期1171-1180,共10页Optics and Precision Engineering

基　　金：国家863高技术研究发展计划资助项目(No.2007AA01Z423);国防"十一五"基础研究资助项目(No.C10020060355);重庆市科技攻关重点资助项目(No.CSTC2007AC2018);重庆市自然科学基金资助项目(No.CSTC2008BB2199)

摘　　要：为了改善降质图像质量,提出一种基于双密度双树复小波变换的局域自适应图像去噪算法。分析了双密度双树复小波变换的原理及特点,推导了双变量收缩函数(BSF)。通过并行使用4个2D双密度离散小波变换,且行和列采用不同的滤波器组,实现了对噪声图像的双密度双树复小波分解。根据小波系数的统计特性以及层内和层间系数的相关性,采用结合局域方差估计的双变量收缩函数对小波系数进行处理,并用收缩后的小波系数重构去噪图像。最后,将该算法用于灰度图像和彩色图像去噪实验。实验结果表明:与噪声图像相比,在噪声方差为30时,经该算法去噪后的图像获得的最高峰值信噪比增益达11.72 dB,平均结构相似度最高增加了2.7倍,复合峰值信噪比增益达11.68 dB。此外,对不同噪声方差下的不同噪声图像,该算法在滤除噪声的同时可保留更多的图像细节,极大地改善了去噪图像的视觉质量。In order to improve the quality of the degraded images, an efficient local adaptive image denoising algorithm based on the Double-density Dual-tree Complex Wavelet Transform （DD-DT CWT） is proposed. The principles and characteristics of the DD-DT CWT are analyzed and a Bivariate Shrinkage Function（BSF） is derivated. Then,the noise image decomposition by the DD-DT CWT is implemented by applying four 2-D Double-density Discrete Wavelet Transform（DD DWT） in parallel and distinct filter sets in the rows and columns. According to the statistical properties of wavelet coefficients and the dependency of inter-level with intra-level coefficients, the BSF with local variance estimation is adopted to process wavelet coefficients and to reconstruct the denoised images by the shrunk wavelet coefficients. Finally, the proposed algorithm is tested on some gray and color noisy images. The experimental results indicate that, compared with the noise images, the Peak Signal-to-Noise Ratio （PSNR） gain of the proposed algorithm has reached 11. 72 dB, Mean Structural Similarity （MSSIM） has been 2.7 times higher than that of noise images and the Composite Peak Signal-to-noise Ratio （CPSNR） reaches 11.68 dB when the noise variance is 30. Meanwhile, the algorithm is more efficient in noise removal and edge reservation for all the noise images with different noise variances, which improves the visual quality of the denoised images.

关 键 词：图像去噪 双密度双树复小波变换 双变量收缩函数 平均结构相似度 复合峰值信噪比 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]