Fig. 1. Structure of convGMM with application to compressive sensing.基于convGMM的压缩测量

Fig. 2. Averaged PSNR of reconstructed images from CIFAR-10 dataset as a function of sampling rate.CIFAR-10图像, 不同算法下恢复图像的PSNR随采样率的变化

Fig. 3. Averaged PSNR of reconstructed images from Caltech 101 dataset as a function of sampling rate.Caltech 101图像, 不同算法下恢复图像的PSNR随采样率的变化

Fig. 4. Reconstructed performance comparison of 12 randomly selected “airplane” images from Caltech 101: (a) Original images; (b) images reconstructed by MMLE-convGMM; (c) images reconstructed by MLE-GMM; (d) images reconstructed by KSVD-YALL1; (e) images reconstructed by DCT-YALL1; (f) images reconstructed by DCT-GAP; (g) images reconstructed by DCT-OMP. All of the sampling rates are 0.4.采样率为0.4时, 12张Caltech 101“飞机”图像在不同算法下的恢复情况　(a)原图像; (b) MMLE-convGMM下的恢复图像; (c) MMLE-GMM下的恢复图像; (d) KSVD-YALL1下的恢复图像; (e) DCT-YALL1下的恢复图像; (f) DCT-GAP下的恢复图像; (g) DCT-OMP下的恢复图像

Fig. 5. Averaged PSNR of reconstructed images from CelebA dataset as a function of sampling rate by MMLE-convGMM.MMLE-convGMM算法恢复CelebA图像的PSNR随采样率的变化

Fig. 6. Reconstructed performance of randomly selected CelebA face images: (a) Original images; (b) images reconstructed by MMLE-convGMM. The sampling rates are 0.4.随机选取的CelebA图像的恢复情形　(a)原图像; (b) MMLE-convGMM算法恢复的图像, 采样率为0.4