Single Image Super Resolution using a Joint GMM Method

Abstract

Single image super-resolution (SR) algorithms based on joint dictionaries and sparse representations of image patches have received significant attention in the literature and deliver the state-of-the-art results. Recently, Gaussian mixture models (GMMs) have emerged as favored prior for natural image patches in various image restoration problems. In this paper, we approach the single image SR problem by using a joint GMM learnt from concatenated vectors of high and low resolution patches sampled from a large database of pairs of high resolution and the corresponding low resolution images. Covariance matrices of the learnt Gaussian models capture the inherent correlations between high and low resolution patches, which are utilized for inferring high resolution patches from given low resolution patches. The proposed joint GMM method can be interpreted as the GMM analogue of joint dictionary-based algorithms for single image SR. We study the performance of the proposed joint GMM method by comparing with various competing algorithms for single image SR. Our experiments on various natural images demonstrate the competitive performance obtained by the proposed method at low computational cost.

Maximizing Nonlocal Self‐Similarity Prior for Single Image Super‐Resolution

“…From the above experimental results, we can find that our algorithm has certain advantages in the quality of reconstructed image compared with other state-of-the-art algorithms, and the consuming time is at the intermediate level. By literature retrieval, we noticed that there are some formal resemblances between the algorithms [32,33,38,40,42] and our proposed, but they are different in nature. The algorithms [32,38,40,42] achieved the task of image restoration with the help of GMM.…”

“…By literature retrieval, we noticed that there are some formal resemblances between the algorithms [32,33,38,40,42] and our proposed, but they are different in nature. The algorithms [32,38,40,42] achieved the task of image restoration with the help of GMM. For an image patch to be restored, the assumption that this image patch is generated by just some Gaussian component with highest probability is necessary, and then to estimate the clear version of this patch by wiener filtering.…”

“…Recently, GMM has been a popular and valid tool used in image processing [38] due to the fact that it can smoothly approximate the density distribution of arbitrary shape and is characterized as remarkable efficiency. Specifically, this property shows excellent performance in image modeling based on image patches [39,40]. In this paper, for training set { ,…”

Image denoising using group sparsity residual and external nonlocal self-similarity prior

“…Then, we assume that the k-th Gaussian component is selected for the group Yi. Actually, GMM model is equivalent to the block sparse estimation with a block dictionary having K blocks wherein each block corresponds to the PCA basis of one of the Gaussian components in the mixture [9,18]. Thus, the covariance matrix of the k-th Gaussian component is denoted by Σ k .…”

Super resolution reconstruction of image gradient profile sharpness

“…They obtain similar or improved quality and one or two order of magnitude speed improvements. Palakkattillam Sandeep et al [11] In this paper, they present an Approach the single image SR problem by using a joint Gaussian mixture model learnt from concatenated vectors of high and low resolution patches sampled from a large database of pair of high resolution and the corresponding low resolution image. Covariance matrices of the learnt Gaussian model capture the inherent correlation between high and low resolution patches, which are utilized for inferring high resolution patch from given low resolution patch.…”