Image super‐resolution based on adaptive cosparse regularisation

Abstract

A novel regularised image super-resolution algorithm is proposed, building on the emerging cosparse or analysis sparse prior models, which are important complementary alternatives to the widely used synthesis sparse counterpart. Moreover, to achieve adaptivity to the varying local structures of natural images, the patch space is partitioned into meaningful subspaces by clustering and learn analysis sub-dictionary for each cluster are partitioned, which are performed online and iteratively based solely on the current available image information, for maximum generality and flexibility. In addition, non-local feature self-similarity is incorporated for further reconstruction quality enhancement. Experimental results show that the proposed approach gives favourable results with respect to the state-of-the-art methods.

Image Super-Resolution Reconstruction Using Adaptive Co-sparse Regularization with Dual Dictionary

“…4(b), that the bicubic method results contained jaggy artefacts and produce degraded SR image, and are unable to overcome the deblurring problem. Chen [22], overcome the limitation of bicubic method by providing better algorithm which leads to be satisfactory SR reconstruction that contained a better visual quality but having less texture detailed compare with proposed method. Gou [21] achieved better results and outperforms previous method in terms of visual resolution but cannot maintain the smoothing edges of reconstructed SR image.…”

“…Fig. 4(a) shows the original HR image, 4(b) shows the bicubic interpolation method, and 4(c)-(e) show the results of various algorithm Chen [22], Gou method [21], and the proposed method respectively. It can be seen from Fig.…”

“…It can be seen that bicubic method achieved low PSNR in all of applied images. Chen [22] algorithm can generated SR image and achieved better PSNR results than the bicubic method. Gou [22] approach achieved better PSNR because of applied external dictionary which is constructed from high resolution images and that leads to be good reconstructed SR image.…”

“…New method has been applied recently to achieve the goal of an SR image by externally applied high resolution dictionary to reconstruct an image, the applied dictionary is not selected adaptively and processing of each patch required large amount of calculation, consequently leads to be large time consuming algorithm [21]. Chen et al [22] introduced an adaptive cosparse method to reconstruct a high resolution HR image, patch is divided under the subspace domain by cluster form and then learned the subdictionary from cluster respectively, and the resulting SR image is with less texture detailed due to missing HF component in applied dictionary. Recently, SR can be reconstruct based on clustering algorithm by using sparse algorithm, the dictionary is used in reconstruction phase which is unable to produce a satisfactory SR reconstruction [23].…”

Super-Resolution Image Reconstruction with Improved Sparse Representation

“…Huahua et. al., [13] proposed a new technique based on adaptive co-sparse regularization to improve the SR efficiency by learning sub-dictionary online from the partitioned cluster which can produce good reconstruction but required large amount of memory. Yincheng et.…”