Fig 1 - uploaded by Jian Pu
Content may be subject to copyright.
Matrix-mulitplication-like operation between Xt(xt ij ) m×d and Xs(xs ij ) n×d. We follow the method of matrix multiplication to compute the distance matrix, with the computing"kernel" modified.
Source publication
Learning based super-resolution can recover high resolution image with high quality. However, building an interactive learning
based super-resolution system for general images is extremely challenging. In this paper, we proposed a novel GPU-based Interactive Super-Resolution system through Neighbor Embedding (ISRNE). Random projection tree (RPtree...
Context in source publication
Context 1
... approach is very similar to the matrix multiplication between X t and X T s , except that the computing kernel is modified. We can very conveniently cope with the computation utilizing currently available parallel matrix multiplication algorithms to obtain significant performance improvement. An illustration of this operation is shown in Fig. ...
Similar publications
This paper utilizes data mining algorithms to predict the evaluation value of new items by target users in an interactive learning environment. To enhance data quality, redundant data in the dataset is eliminated. To provide better learning recommendations and personalized services, prediction accuracy is assessed by calculating the deviation betwe...
Deep convolutional neural networks (CNNs) have great improvements for single image super resolution (SISR). However, most of the existing SISR pre-training models can only reconstruct low-resolution (LR) images in a single image, and their upsamling factors cannot be non-integers, which limits their application in practical scenarios. In this paper...
Recently, facial-expression recognition (FER) has primarily focused on images in the wild, including factors such as face occlusion and image blurring, rather than laboratory images. Complex field environments have introduced new challenges to FER. To address these challenges, this study proposes a cross-fusion dual-attention network. The network c...
![Interactiveness detection results on HICO-DET [1] and V-COCO [8].](profile/Yong-Lu-Li/publication/362324240/figure/tbl1/AS:1183052038451206@1659072994817/Interactiveness-detection-results-on-HICO-DET-1-and-V-COCO-8_Q320.jpg)
![Results of ablation studies on HICO-DET [1].](profile/Yong-Lu-Li/publication/362324240/figure/tbl2/AS:1183052038443018@1659072994838/Results-of-ablation-studies-on-HICO-DET-1_Q320.jpg)
Human-Object Interaction (HOI) detection plays a crucial role in activity understanding. Though significant progress has been made, interactiveness learning remains a challenging problem in HOI detection: existing methods usually generate redundant negative H-O pair proposals and fail to effectively extract interactive pairs. Though interactiveness...
Deep convolutional neural networks (CNNs) have great improvements for single image super resolution (SISR). However, most of the existing SISR pre-training models can only reconstruct low-resolution (LR) images in a single image, and their upsamling factors cannot be non-integers, which limits their application in practical scenarios. In this lette...
Citations
... At present, learning based ISR methods are thought as the hot research topics. Currently, typical learning based image ISR methods generally are summarized as samples based ones [19,20], Local Linear Embedding based Manifold Learning (LLE-ML) methods [19], Neighbor Embedding (NE) methods [21,22] , K- Nearest Neighbor (K-NN) methods [21], Kernel ridge regression methods [23], wavelet coefficient dictionary methods [22], and sparse representation based methods [24,25] and so on. Among these typical ones, sparse representation based methods are the most popular and many ones also has been developed in restoring LR image s [24][25][26][27][28][29]. ...
... Currently, typical learning based image ISR methods generally are summarized as samples based ones [19,20], Local Linear Embedding based Manifold Learning (LLE-ML) methods [19], Neighbor Embedding (NE) methods [21,22] , K- Nearest Neighbor (K-NN) methods [21], Kernel ridge regression methods [23], wavelet coefficient dictionary methods [22], and sparse representation based methods [24,25] and so on. Among these typical ones, sparse representation based methods are the most popular and many ones also has been developed in restoring LR image s [24][25][26][27][28][29]. Sparse representation based theories can well solve many inverse problems existing in the fields of images. ...
A new Image Super-resolution Reconstruction (ISR) method combined a modified K-means based Singular Value Decomposition (M_K-SVD) model and Regularized Adaptive Matching Pursuit (RAMP) algorithm is proposed in this paper. In the M_K-SVD model, the maximum sparsity of sparse coefficients is considered. In the condition of the unknown sparsity of the original signals, RAMP algorithm can choose automatically and adaptively the candidate set, and utilize the regularization process to implement the final support set so as to finish accurately the task of signal reconstruction. Combined the advantages of M_K-SVD and RAMP algorithm, for LR images and High Resolution (HR) images, the LR and HR dictionaries are trained. And then, utilized the optimized LR sparse coefficient vectors and the HR dictionary, the HR image patches can be estimated. And considered the original locations of HR image patches to be restored, the LR images can be reconstructed. However, LR images contain much unknown noise, so, before training dictionaries, the LR images are first preprocessed by M_K-SVD model. In test, human-made LR images (i.e. natural images' degraded versions) and real LR images (i.e. millimeter wave images, MMW) are respectively used to testify our method proposed. Further, compared our ISR method with those of the basic K-SVD, Regularized Orthogonal Matching Pursuit (ROMP), RAMP, and Sparsity Adaptive Matching Pursuit (SAMP) and so on, experimental results testified the ISR validity of our method proposed. Meanwhile, the Signal Noise Ratio (SNR) criterion is used to measure restored human-made LR images, and the Relative Signal Noise Ratio (RSNR) criterion is used to test the quality of MMW image restored. Experimental results prove that our method is indeed efficient in the research field of ISR reconstruction.
... In the super-resolution phase, for each LR patch in a test image, the most similar database LR patch is found, and its corresponding database HR patch is used. Improved variants of NN methods were proposed in [4], [6], [7]. ...
Example-based super-resolution (SR) methods learn the correspondences between low resolution (LR) and high-resolution (HR) image patches, where the patches are extracted from a training database. To reconstruct a single LR image into a HR one, each LR image patch is processed by the previously trained model to recover its corresponding HR patch. For this reason, they are computationally inefficient. We propose the use of a selective patch processing technique to carry out the super-resolution step more efficiently, while maintaining the output quality. In this technique, only patches of high variance are processed by the costly reconstruction steps, while the rest of the patches are processed by fast bicubic interpolation. We have applied the proposed improvement on representative example-based SR methods to super-resolve text images. The results show a significant speed up for text SR without a drop in theocrat accuracy. In order to carry out an extensive and solid performance evaluation, we also present a public database of text images for training and testing example-based SR methods.
... [1] presents two such tree structures, each adapting to a different notion of intrinsic dimensionality. Both variants have already found numerous applications in regression [7], spectral clustering [8], face recognition [9] and image super-resolution [10]. ...
The Random Projection Tree structures proposed in [Freund-Dasgupta STOC08]
are space partitioning data structures that automatically adapt to various
notions of intrinsic dimensionality of data. We prove new results for both the
RPTreeMax and the RPTreeMean data structures. Our result for RPTreeMax gives a
near-optimal bound on the number of levels required by this data structure to
reduce the size of its cells by a factor $s \geq 2$. We also prove a packing
lemma for this data structure. Our final result shows that low-dimensional
manifolds have bounded Local Covariance Dimension. As a consequence we show
that RPTreeMean adapts to manifold dimension as well.
CubeSats have the demonstrated potential to contribute to commercial, scientific, and government applications in remote sensing, communications, navigation, and research. Despite significant research into improving CubeSat operational efficiency, there remains one fundamental limitation of CubeSats for EO imaging applications: the small lenses and short focal lengths result in imagery with low spatial resolution. This paper reviews the previous research on super-resolution techniques and proposes potential applications of super-resolution to CubeSat EO imagery.
Face super-resolution is to synthesize a high resolution facial image from a low resolution input, which can significantly improve the recognition for computer and human. Regularization plays a vital role in ill-posed problems. The use of examples becomes much more effective when handling narrow family of images, such as face images. A properly chosen regularization can direct the solution toward a better quality outcome. An emerging powerful regularization is one that leans on image examples. This paper proposed a face hallucination method using example-based regularization. The work is specially targeted at improving the quality of high magnification. Our work follows the pyramid framework and assigns several high-quality candidate patches for each location in the degraded image. All problematic examples are rejected by defining an error function which embodies the example-based regularization. After repeated pruning, the reconstruction is done when there is only one candidate patch left in each location. The encouraging experimental results provide some hints that our approach is effective.
We explore in this paper an efficient algorithmic solution to single image super-resolution (SR). We propose the gCLSR, namely graph-Constrained Least Squares Regression, to super-resolve a high-resolution (HR) image from a single low-resolution (LR) observation. The basic idea of gCLSR is to learn a projection matrix mapping the LR image patch to the HR image patch space while preserving the intrinsic geometric structure of the original HR image patch manifold. Even if gCLSR resembles other manifold learning-based SR methods in preserving the local geometric structure of HR and LR image patch manifolds, the innovation of gCLSR lies in that it preserves the intrinsic geometric structure of the original HR image patch manifold rather than the LR image patch manifold, which may be contaminated by image degeneration (e.g., blurring, down-sampling and noise). Upon acquiring the projection matrix, the target HR image can be simply super-resolved from a single LR image without the need of HR-LR training pairs, which favors resource-limited applications. Experiments on images from the public database show that gCLSR method can achieve competitive quality as state-of-the-art methods, while gCLSR is much more efficient in computation than some state-of-the-art methods.
An optimal recovery based neural-network Super Resolution algorithm is developed. The proposed method is computationally less
expensive and outputs images with high subjective quality, compared with previous neural-network or optimal recovery algorithms.
It is evaluated on classical SR test images with both generic and specialized training sets, and compared with other state-of-the-art
methods. Results show that our algorithm is among the state-of-the-art, both in quality and efficiency.
In this paper we study the usefulness of different local and global, learning-based, single-frame image super-resolution reconstruction
techniques in handling three specific tasks, namely, de-blurring, de-noising and alias removal. We start with the global,
iterative Papoulis–Gerchberg method for super-resolving a scene. Next we describe a PCA-based global method which faithfully
reproduces a super-resolved image from a blurred and noisy low resolution input. We also study several multi-resolution processing
schemes for super-resolution where the best edges are learned locally from an image database. We show that the PCA-based global
method is efficient in handling blur and noise in the data. The local methods are adept in capturing the edges properly. However,
both local and global approaches cannot properly handle the aliasing present in the low resolution observation. Hence we propose
an alias removal technique by designing an alias-free upsampling scheme. Here the unknown high frequency components of the
given partially aliased (low resolution) image is generated by minimizing the total variation of the interpolant subject to
the constraint that part of alias free spectral components in the low resolution observation are known precisely and under
the assumption of sparsity in the data.
In recent years, a series of manifold learning algorithms have been proposed for nonlinear dimensionality reduction (NLDR).
Most of them can run in a batch mode for a set of given data points, but lack a mechanism to deal with new data points. Here
we propose an extension approach, i.e., embedding new data points into the previously-learned manifold. The core idea of our
approach is to propagate the known coordinates to each of the new data points. We first formulate this task as a quadratic
programming, and then develop an iterative algorithm for coordinate propagation. Smoothing splines are used to yield an initial
coordinate for each new data point, according to their local geometrical relations. Experimental results illustrate the validity
of our approach.
