A non-stationary image prior combination in super-resolution

A l1 Norm Based Image Prior Combination in Multiframe Superresolution

“…This section presents experimental results of our proposed method on simulated and real images and compares it with (1) bicubic interpolation (BBC), that is, bicubic interpolation of the LR reference frame, (2) the variational SR method using simultaneous autoregressive model (SAR) prior [16] (denoted by SAR), (3) the variational SR method using TV prior model [16] (denoted by TV), and (4) the nonstationary image prior combination method [17] with the filter combination NF3, because it has the best performance in the combination selection (denoted by NS NF3). It is easy to see that the 1 prior method in [20] is a special case of our proposed method.…”

“…For the sake of comparison, the filters in Section 2.2 and the filter combinations are mainly designed as the ones in [17] except the Haar filters. The image filters used in this paper are the first-order difference (f.o.d.)…”

“…In [21], the combination of the sparse prior and nonsparse SAR prior was proposed, but the combination coefficient had to be determined by the hand empirically. In [17], the author suggested a nonstationary image prior combination method based on a general combination of spatially adaptive image filters, and the combination coefficient can be estimated automatically. However, this method favors oversmooth image estimates because of the heavy penalty of the 2 norm in the prior model.…”

“…Therefore, methods in spatial domain have become more popular in recent years due to their convenient incorporation of prior knowledge. These methods fall into two categories: deterministic approach, such as projection onto convex sets (POCS) [9] and iterative back projection (IBP) [10]; stochastic approach, including maximum likelihood (ML) [11], maximum a posterior (MAP) [12][13][14], and Bayesian method [15][16][17]. Compared with the deterministic approach, the stochastic approach has more flexibility in modeling the imaging formation process and the prior knowledge [5], and using the hierarchical Bayesian method, both the high resolution image and model parameters can be estimated simultaneously and automatically.…”

A Generalized Detail-Preserving Super-Resolution method

“…The focus of this paper is the multi-frame SR method. Recently, the Bayesian method [7][8][9] has become the most widely used method, which mainly contains the likelihood function and image prior function. The likelihood function describes the difference between the observed LR images and the HR image, and the image prior function abstractly expresses the prior knowledge of the HR image.…”