摘要
基于小波系数先验模型的图像处理方法是现代图像处理技术的重要理论框架之一。针对传统的高斯型或拉普拉斯型先验模型的统计描述精度缺陷,本文提出利用贝叶斯(Bayesian)神经网络模型对图像小波系数的统计特性进行精确建模,结合现代粒子(或Montel Carlo:MC)采样技术——Gibbs采样进行模型参数的估计,并重点考察了各尺度下的正交小波系数先验信息的建模过程,最后利用先验模型图像处理框架,实现图像的噪声抑制。仿真模拟结果表明:一方面,基于贝叶斯神经网络的小波先验模型建模准确,较好地描述了各尺度小波系数的先验信息;另一方面,从图像去噪性能来看,基于建议先验模型的图像质量在客观指标和主观视觉上都有显著的提高。
Image processing based wavelet coefficients prior statistical models plays one of great improtant roles in modern image processing techniques. Owing to the defaults of fitting of Gaussian or Laplace functions, a Bayesian model of neural network（BMNN） to study the statistical dependency of wavelet coefficients is firstly presented. Secondly, its parameters are estimated by modern particle samplers （Monte Carlo） methods--Gibbs algorithm according to the characteristics of the suggested BMNN model. Then the relation- ship of wavelet coefficients is discussed in detail. Finally, a practical application of denoising image by using the BMNN model is demonstrated and the result shows that, on one hand the suggested method can express wavelet coefficients dependency efficiently, on the other, high quality visual effects and peak signal- to-noise ratio （PSNR） are achieved.
出处
《重庆师范大学学报（自然科学版）》
CAS
2009年第3期65-68,共4页
Journal of Chongqing Normal University:Natural Science
基金
重庆市教委科学技术项目(No.KJ090829)
重庆师范大学青年基金项目(No.08XLS13)
