小波树结构在贝叶斯压缩感知图像重构中的应用研究  

Research on Application of Wavelet Tree Structure in Bayesian Compressive Sensing Image Reconstruction

作　　者：袁琴[1,2] 吴宣够[2] 熊焰[2] 

机构地区：[1]黄山学院机电与信息工程学院,黄山245041 [2]中国科学技术大学计算机科学与技术学院,合肥230026

出　　处：《计算机科学》2014年第3期314-318,F0003,共6页Computer Science

基　　金：国家自然科学基金:未来无线网络体系结构与关键技术(61170233);校级科研项目(2010xkj005);教育厅科研项目(KJ2011B164)资助

摘　　要：结合贝叶斯和压缩感知理论,提出了一种基于小波变换的图像压缩和重建方法。这种算法充分利用了小波变换系数的结构特征和相关性,有效地提高了图像的压缩比例和重建精度。对小波变换的尺度系数采用基于预测的恢复算法;对高频系数的恢复结合了贝叶斯理论和压缩感知理论,采用了一种基于回归模型的方法,通过高斯混合参数对未知权值参数赋予确定的先验分布,以限制系数的稀疏性。该方法能够得到未知参数的一组具有较高概率的模型,从而实现系数在MMSE意义下的重建。与现有的图像压缩方法以及其它基于压缩感知的图像压缩方法相比,该算法能够获得较高的图像重建质量和较大的图像压缩比。Combining Bayesian learning and compressive sensing,an image compression and reconstruction method based on wavelet transform was proposed in this article.Utilizing the specific structure and correlation of wavelet transforming coefficients,this method improves the image compression rate and reconstruction accuracy effectively.At same time,a regression model based on prediction is adopted in coefficient reconstruction.Gaussian mixture parameters are used to predefine the prior conditional density of the unknown parameters in order to enforce the sparsity.This method can get a group of model with high probability of the coefficients,and result in reconstruction of the image in sense of MMSE.Compared with other image compression methods and CS based image reconstruction methods,the proposed method can get reconstruction images with high quality and get bigger compression rate.

关 键 词：压缩感知 小波变换 图像压缩 贝叶斯 

分 类 号：TN95[电子电信—信号与信息处理]