Soft-decision adaptive interpolation (SAI) provides a powerful framework for image interpolation. The robustness of SAI can be further improved by using weighted least-squares estimation, instead estimation in both the parameter and data steps. To address mismatch issue "geometric duality" during residuals (prediction errors) are according to geometric similarity between pixel interest residuals. modeling weights with well-known bilateral filter. Experimental results show that there is...

In this study, the authors propose a new image interpolation technique using bilateral filter to estimate unknown high-resolution pixels. Compared with least-squares estimation, small-kernel has advantages of fast computation and stability. The range distance is estimated novel maximum posterior in order consider both diagonal vertical–horizontal correlations. For consideration global consistency, pixel-based soft-decision estimation (SAI) proposed constrain consistency edge statistic within...

In recent years, deep convolutional neural networks (CNNs) have been widely used for image super-resolution (SR) to achieve a range of sophisticated performances. Despite the significant advancement made in CNNs, it is still difficult apply CNNs practical SR applications due enormous computations convolutions. this paper, we propose two lightweight using depthwise separable convolution real-time SR. Specifically, divides standard into and pointwise significantly reduce number model...

Recently, there have been significant advances in image super-resolution based on generative adversarial networks (GANs) to achieve breakthroughs generating more images with high subjective quality. However, are remaining challenges needs be met, such as simultaneously recovering the finer texture details for large upscaling factors and mitigating geometric transformation effects. In this paper, we propose a novel robust GAN (i.e. namely RSR-GAN) which can perform both details. Specifically,...

Thanks to the efficient retrieval speed and low storage consumption, learning hash has been widely used in visual tasks. However, known hashing methods assume that query samples lie homogeneous feature space within same domain. As a result, they cannot be directly applied heterogeneous cross-domain retrieval. In this article, we propose generalized image transfer (GITR) problem, which encounters two crucial bottlenecks: 1) may come from different domains, leading an inevitable domain...

New edge-directed interpolation (NEDI) consists two steps. The steps are parameter and data estimation. second step can be replaced by a recently proposed technique called soft-decision to consider the consistency of image structure during this original idea both is assume equal variances for all estimation errors, such that an ordinary least squares (OLS) estimator used. Due existence noise, different object layers, changing in structures, spatial distance missing data, etc, we observe...

Image up-sampling in the discrete cosine transform (DCT) domain is a challenging problem because DCT coefficients are de-correlated, such that it nontrivial to estimate directly high-frequency from observed low-frequency coefficients. In literature, DCT-based algorithms usually pad zeros as or with limited success mainly due nonadaptive estimator and restricted information single image. this paper, we tackle of estimating spatial by proposing learning-based scheme using an adaptive k-nearest...

In this paper, we propose an iterative Wiener filter which can simultaneously perform interpolation and restoration by using non-local means to directly model the correlation between desired high-resolution image observed low-resolution image. A novel mechanism is proposed control decay speed of function while iteratively updating both estimated During iterations, decomposed into patches with similar intensities at initial iterations are connected naturally good convergence. Experimental...

A typical dynamic reconstruction-based super-resolution video involves three independent processes: registration, fusion and restoration. Fast systems apply translational motion compensation model for registration with low computational cost. Traditional assumes that the whole spectrum of pixels is consistent between frames. In reality, frequency component often varies significantly. We propose a via classification systems. novel idea to implement by combining up-sampled current frame high...

The hybrid DCT-Wiener-based (DCT-WB) interpolation scheme provides a powerful framework to interpolate an image by utilizing the information in both spatial and DCT domain. In this paper, we investigate bottleneck of propose 2D non-separable block-based Wiener filter for scheme. is learnt using training pairs through minimum mean squares error estimation. proposed resolves quarter-pixel shift issue much better performance over original 1D 6-tap pixel-based filter. Experimental results show...

In general, existing research on single image super-resolution does not consider the practical application that, when transmission is over noisy channels, effect of any possible geometric transformations could incur significant quality loss and distortions. To address this problem, we present a new robust method in paper, where spatially-transformed deep learning framework established to simultaneously perform both transformation super-resolution. The proposed seamlessly integrates residual...

Video super-resolution aims to increase the resolution of videos by exploiting intra-frame and inter-frame dependencies low-resolution video sequences. There are usually two dependent steps in super-resolution: motion compensation reconstruction. In this paper, we propose a new deep learning framework without explicit estimation utilizing self-attention model exploit full receptive field input frames. other words, proposed neural network extracts local features at all spatial-temporal...

As deep learning neural networks (DNNs) advance and increase in computational complexity, particularly terms of memory cost, it becomes difficult to implement DNNs fixed-point memory-sparse environments (e.g. integrated circuits consumer electronics). Thus, the training must be reformulated balance hardware costs needed represent often millions parameter weights such machine models. This paper proposes a novel optimization approach that simultaneously minimizes complexity (total memory)...

Recently, the probabilistic motion field was proposed for super-resolution reconstruction (SRR). In interpolation step of SRR, a missing pixel can be estimated by weighted average neighboring pixels, which are errors with pixel. However, far from true values due to approximated in calculating errors. Hence, this paper, we propose linear model better approximate errors, results quality. Experimental show that gain 0.6 dB PSNR is achievable using model, and only small number pixels have used...

In this paper, we propose two approaches to generate a high resolution (HR) image from low (LR) version. It is commonly referred as single-image super-resolution(SISR). Our are inspired by the spatial transformer (ST) module and Very Deep Convolutional Network (VDSR). The neural network originally used for geometric transformations of images, while VDSR super-resolution. first approach, add ST with HR images. use transform makes more robust transformations. We second approach replace...

In novel view synthesis using the depth image based rendering, there exist some unknown pixel intensities (holes) due to unexposed area. this paper, we propose a depth-assisted nonlocal means algorithm fill holes information in current frame and other frames synthesized video. The has been successfully applied video super-resolution applications. hole filling, challenge is irregular interpolation. However, can be relatively reliable map, such that are able combine intensity information, make...

For many practical applications, it is essential to address both geometric corrections and missing information reconstruction of face images natural images. However, unfavorable separate the problem into two sub-tasks due error accumulations sequential tasks. In this paper, we propose a novel robust framework via deep feature transformations simultaneously image completion. Specifically, our proposed realizes multiple channel spatial tackle corrections, completion through non-linear features...

In this paper, we propose a novel image interpolation technique using the small-kernel bilateral filter. The range distance of filter is estimated maximum posterior (MAP) estimation, in order to consider both diagonal and vertical-horizontal correlations image. proposed method requires few arithmetic operations, such that it suitable for real time zoom, video up-scaling transcoding applications HDTV systems. Subjective evaluation shows approach outperforms conventional terms edge sharpness...