The distribution model of sparse signals is a key influencing factor of the effectiveness of Com-pressed Sensing( CS) signal reconstruction based on Approximate Message Passing( AMP) . In actual ima-ges,there are sharp differences in the model parameters of wavelet approximation coefficients,horizontal de-tail coefficient,vertical detail coefficients and the diagonal detail coefficients at all levels. And the current AMP method is based on Laplace model,Bernoulli Gaussian( BG) and Gaussian Mixture,which,however, fails to take such differences into consideration. Therefore,the reconstruction results will be affected. In order to improve the accuracy of model estimation,this paper estimates the BG model parameters of wavelet coefficients at all levels respectively, and on this basis, proposes an improved method called Expectation Maximization Separately Segment Bernoulli Gaussian Approximate message passing( EM-SSBG-AMP) for image CS sparse reconstruction. Simulation results show that, under the same sampling rate, the peak signal-to-noise ratio( PSNR) of the new algorithm is obviously higher than that of the 5-order Expectation Maximization Gaussian Mixture Approximate message passing( EM-GM-AMP) and the reconstruction time is similar to that of 5-order EM-GM-AMP.%稀疏信号的分布模型是影响基于近似信息传递( AMP)的压缩感知( CS)信号重建效果的关键因素。因实际图像的小波近似系数、各级的水平细节系数、垂直细节系数以及对角细节系数的模型参数存在较大差异，现有基于拉普拉斯、贝努力高斯( BG)和高斯混合等模型的AMP方法因未考虑此差异而影响重建效果。为了提高模型估计的准确性，将各级小波系数的BG模型参数分开估计，进而提出了一种改进的图像压缩感知稀疏重建的新方法，即期望最大分段贝努力高斯近似信息传递算法( EM-SSBG-AMP)。仿真结果表明，相同采样率下，新算法的峰值信噪比( PSNR)明显高于5阶期望最大高斯混合近似信息传递算法( EM-GM-AMP)，重建时间与5阶EM-GM-AMP相当。
展开▼
