Super-resolution is a process that combines information from some low-resolution images in order to produce an image with higher resolution. In most of the previous related work, the blurriness that is associated with low resolution images is assumed to be due to the integral effect of the acquisition device’ s image sensor. However, in practice there are other sources of blurriness as well, including atmospheric and motion blur that may be applied to low resolution images. The research done in this paper provides a super-resolution image from some low-resolution images suffering from blurriness due to defocus. In contrast to motion blur kernels that are sparse, the defocus blur kernel is non-sparse and continuous. Because of the continuity property of defocus blurring kernel, in this paper, we bound the gradient of blurring kernel using proper regularizers to satisfy this property. Experimental results on synthetic data demonstrate the effectiveness of the proposed method to produce high resolution and de-blurred images from some blurry low-resolution images.

The goal of multi-frame Super Resolution (SR) is to fuse multiple Low Resolution (LR) images to produce one High Resolution (HR) image. The major challenge of classic SR approaches is accurate motion estimation between the frames. To handle this challenge, fuzzy motion estimation method has been proposed that replaces value of each pixel using the weighted averaging all its neighboring pixels in all LR images which carries the degree of similarity between image blocks centered on two pixels. Since in case of rotation between LR images, comparing the gray level of blocks around the pixels is not a suitable criterion for calculating weight, so, magnitude of Zernike Moments (ZM) has been used as a rotation invariant feature. Due to the lower sensitivity of Pseudo Zernike Moments (PZM) to noise and the higher discrimination capability of it for the same order compared to ZM, in this paper, we propose a new method based on magnitude of PZM of the blocks as a rotation invariant descriptor for representation of pixels in weight calculation. Experimental results on several image sequences show that the performance of the proposed algorithm is better than the existing and new techniques from the aspect of PSNR and visual image quality.

Super-resolution (SR) aims to overcome the ill-posed conditions of image acquisition. SR facilitates scene recognition from low-resolution image(s). Various approaches have tried to aggregate the informative details of multiple low-resolution images into a high-resolution one. In this paper, we present a new robust fuzzy super resolution approach. Our approach, firstly registers two input image using SIFT-BP-RANSAC registration. Secondly, due to the importance of information gain ratio in the SR outcomes, the fuzzy regularization scheme uses the prior knowledge about the low-resolution image to add the amount of lost details of the input images to the registered one using the common linear observation model. Due to this fact, our approach iteratively tries to make a prediction of the high-resolution image based on the predefined regularization rules. Afterwards the low-resolution image have made out of the new high-resolution image. Minimizing the difference between the resulted low-resolution image and the input low-resolution image will justify our regularization rules. Flexible characteristics of fuzzy regularization adaptively behave on edges, detailed segments, and flat regions of local segments within the image. General information gain ratio also should grow during the regularization. Our fuzzy regularization indicates independence from the acquisition model. Consequently, robustness of our method on different ill-posed capturing conditions and against registration error noise compensates the shortcomings of same regularization approaches in the literature. Our final results show reduced aliasing achievements in comparison with similar recent state of the art works.

Hyperspectral images have high spectral resolution. But, due to the tradeoff between spectral and spatial resolution and various hardware constraints, imaging a hyperspectral image with high spatial resolution is not practical. Hyperspectral super resolution is a soft approach to solve this challenge. Recently, deep learning based methods such as convolutional neural network (CNN) show great success in this field. But, the contextual details in object boundaries and anomalies present in the scene are not well addressed. To this end, a new CNN based framework is proposed for hyperspectral image super resolution in this work. To improve ability of the convolutional blocks in simultaneous extraction of spectral and spatial characteristics, the weighted Gabor features are concatenated in output of the defined convolutional blocks. To extract more details containing anomalous targets present in the scene, the anomaly scores of pixels are calculated and used for weighting the Gabor features. The experiments on three real hyperspectral images acquired by AVIRIS and ROSIS sensors show superior performance of the proposed framework compared to several state-of-the-art methods based on CNN and residual networks. In addition to common super resolution metrics such as SAM and ERGAS, the efficiency of different methods are evaluated according to the classification accuracy metrics such as overall accuracy and kappa coefficient. The overall classification accuracy is increased from 70.39 to 88.23 in Indian dataset, from 86.07 to 96.20 in Pavia University dataset, and from 95.82 to 99.12 in Pavia center dataset.