Self-Learning of Edge-Preserving Single Image Super-Resolution via Contourlet Transform

Size: px
Start display at page:

Download "Self-Learning of Edge-Preserving Single Image Super-Resolution via Contourlet Transform"

Transcription

1 Self-Learning of Edge-Preserving Single Image Super-Resolution via Contourlet Transform Min-Chun Yang, De-An Huang, Chih-Yun Tsai, and Yu-Chiang Frank Wang Dept. Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan Dept. Electrical Engineering, National Taiwan University, Taipei, Taiwan Research Center for Information Technology Innovation, Academia Sinica, Taipei, Taiwan {d , b979000, Abstract We present a self-learning approach for single image super-resolution (SR), with the ability to preserve high frequency components such as edges in resulting highresolution (HR) images. Given a low-resolution (LR) input image, we construct its image pyramid and produce a superpixel dataset. By extracting context information from the super-pixels, we propose to deploy context-specific contourlet transform on them in order to model the relationship (via support vector regression) between the input patches and their associated directional high-frequency responses. These learned models are applied to predict the SR output with satisfactory quality. Unlike prior learning-based SR methods, our approach advances a self-learning technique and does not require the selfsimilarity of image patches within or across image scales. More importantly, we do not need to collect training LR/HR image data in advance and only require a single LR input image. Empirical results verify the effectiveness of our approach, which quantitatively and qualitatively outperforms existing interpolation or learning-based SR methods. Keywords-Super-resolution, self-learning, contourlet transform I. INTRODUCTION Super-resolution (SR) aims at synthesizing a highresolution (HR) image from its multiple or single lowresolution (LR) versions. Generally, SR approaches can be classified into two categories: reconstruction and learning (or example) based methods. Reconstruction-based SR approaches require image patches from several LR images (or within a single LR image) for constructing the SR output. As can be expected, proper registration and alignment techniques need to be performed for this type of SR methods [], [2]. Learning (or example) based SR approaches typically search for similar patches from training LR image data, and the corresponding HR patches are utilized to synthesize the final SR output (e.g., [3], [4]). Moreover, several works have been proposed to learn the relationship between LR/HR image pairs, and thus the observed models can be used for predicting the SR version from the input LR image (e.g., [5], [6], [9], [0], [7]). We note that, in [5], [7], contourlet transform has been applied to exploit directional high frequency components presented in training images. However, since the above learning based methods need to collect proper training LR/HR image data beforehand, the learned SR models and the associated performance will be sensitive to their choice of training data. As a result, these approaches might not generalize well to real-world SR applications. Recently, self-learning based approaches [2], [3], [4] have been proposed to synthesize the SR image using information extracted from a single LR input image. Glasner et al. [2] combines both classical and examplebased SR techniques, which search for similar patches within and across image scales to produce the SR output. Yang et al. extends this idea and learns a joint sparse coding dictionary for LR/HR images to obtain a refined result. Different from [2], Wang et al. [4] does not search for similar patches. Instead, they advocate the learning of context-specific information across images patches and use the observed models to predict the HR output. In this paper, we propose a novel self-learning approach for single image SR. To overcome existing SR approaches with blurred results, we utilize contourlet transform [8] and aim at preserving high frequency components in the SR output. In our proposed framework, we extract and learn context-specific information from a LR input and its image pyramid. We model the relationship between image patches of the same context category in terms of their directional frequency components. Once this learning process is complete, each patch of the input image will be used to predict the final SR output using the associated models. Different from prior SR methods considering contourlet transform [5], [7] or most learning-based SR methods, we do not require the collection of any training LR/HR image data in advance and we do not assume any specific image priors. With the proposed self-learning processing, we can produce satisfactory SR results given a single input LR image. Moreover, based on the use of contourlets for edge-preserving SR, we further integrate multi-processing techniques with a multicore PC, which results in very impressive processing time compared to existing learning-based SR methods. In the following sections, we will detail our proposed method, followed by quantitative and qualitative experimental results and comparisons with state-of-the-art learning-based SR approaches.

2 DFB Decomposition Contourlet Parallel Processing I SR I 0 U i U + H + H i I - C i- 2 C i- 3 C 4 i- C i- 2 Context Category I (a) (b) (c) (d) (e) Figure. Flowchart of our SR framework. (a) Input image I 0 and its lower-resolution versions (e.g., I, etc.). (b) Extraction and clustering of image segments in terms of their context information.(c) Deriving the up-sampled version U i from I i via bicubic interpolation, the associated high frequency component image H i, and four response images of directional frequency components {C j i, j =,..., 4} (Section II-B). (d) Learning of directional frequency and context specific models for SR with parallel processing techniques (Section II-B2). (e) Refining U + into I SR with H + using predicted {C j 0, j =,..., 4}. II. CONTEXT-AWARE SINGLE IMAGE SUPER-RESOLUTION WITH CONTOURLETS A. Context-Constrained Super-Pixel Categorization Inspired by Glasner et al. [2], who constructs an image pyramid from a single LR image input and searches for similar patches for SR, we advance this framework in our proposed method. Different from [2], we do not require the reoccurrence of image patches within or across scales. Instead, we focus on learning context information from such an image pyramid for edge-preserving SR, while no additional training LR/HR image data is needed. We now explain how we build the image pyramid from a single LR input image, and how we collect a super-pixel database from this pyramid for each context category. Given a LR image, we downgrade its resolution by bicubic downsampling to form images with different resolutions {I i }, as shown in Figure (a). Note that the resolution of each down-scaled image I i is a quarter of that of corresponding higher-resolution version I i+. To determine the context information for each image segment in this pyramid, we apply mean-shift [6] to over-segment each {I i } into superpixels. To categorize these super-pixels into different context categories, we extract the textural feature for each superpixel by calculating the responses of derivative filter banks with 6 different orientations and at 3 different scales (as suggested in [5]). The response of each filter is quantized into a histogram with 20 bins, and we obtain the final textural feature as a = 360 dimensional vector. Since the number of context categories for an image cannot be known a priori, we apply affinity propagation (AP) [7] to perform unsupervised clustering on the above super-pixels using their textural features. We note that, since AP divides the data instances into distinct groups simply by calculating the similarity between instances, it is an automatic and unsupervised clustering algorithm without the need to specify the number of clusters beforehand. In our proposed framework, we apply the χ 2 -distance to measure the difference between each super-pixel pair S i and S j, as suggested in [4]. AP thus determines the optimal clustering by maximizing the net-similarity N S between super-pixels: N N NS = c ijs(s i, S j ) () i= j= N N N N γ ( c ii)( c ij) γ ( c ij). i= j= In (), N is the total number of super-pixels in the image pyramid I i, s(s i, S j ) = exp( χ 2 (S i, S j )) measures the similarity between super-pixels S i and S j. The coefficient c ij = indicates that the super-pixel S i is the exemplar (i.e., the cluster representative) of the super-pixel S j, and thus S j is categorized to cluster i (and c ii equals since S i itself is the exemplar cluster i). While the first term in () calculates the similarity between super-pixels within each cluster, the second term penalizes the case when superpixels are assigned to an empty cluster (i.e. c ii = 0 but with N j= c ij ). On the other hand, the third term in () penalizes the condition when super-pixels belong to more than one cluster. In practice, γ is set to + to avoid the aforementioned problems. Once this clustering process is complete, as shown in Figure (b), we obtain the context label information for super-pixels extracted from the image pyramid I i. Note that image patches bounded by red rectangles in Figure (b) indicate the exemplars (representatives) of different context categories. i= j= B. Edge-Preserving SR via Contourlet Transform ) Modeling Directional High-Frequency Components with Contourlets: Since the human visual system is sensitive to the high frequency components of images, our proposed self-learning algorithm aims at recovering high-frequency components for the SR output. Among different techniques which extract edges, etc. high-frequency components from images, the contourlet transform [8] has a unique property of decomposing multi-resolution and multi-directional frequency components for image reconstruction. Compared to the wavelet transform which captures only edges present in

3 horizontal, vertical and diagonal directions only, the contourlet transform provides a higher degree of directionality and anisotropy by utilizing basis functions oriented at more directions and with more flexibility. As a result, it is able to preserve the geometric structure (e.g., edges and higherfrequency components) of an image. We now detail how we apply the contourlet transform to extract context-specific and directional high-frequency components from a given LR input image. From Section II-A, we construct a super-pixel database from the input image pyramid I i, and each super-pixel belongs to a specific context category. Take Figure 2 for example, for each lowerresolution image I i in the image pyramid, we upsample its resolution by bicubic interpolation and obtain a higherresolution version U i, which is of the same size as the image I i in the pyramid. Since the image U i is upsampled from I i, it misses some high-frequency details and thus is blurred. We subtract U i from the image I i in the pyramid, and the resulting image H i would correspond to the highfrequency components to be recovered from I i (when using bicubic interpolation). In order to model these highfrequency components, we apply a two-level directional filter bank (DFB) decomposition with 2 2 = 4 subbands on the image H i to produce a stack of four directional components {C j i, j =,..., 4} (also known as contourlets). More specifically, given the input high-frequency image H i, the four contourlets obtained from the DFB decomposition are calculated as: C j i = downsample(h i E j, S j ), j=,..., 4. (2) where E j is a fan-shaped filter bank with non-zero values in horizontal or vertical directions, and the downsample function downgrades the resolution of the convolution output H i E j using a diagonal matrix S j (see [8] for more details). Since the convolution and downsampling processes are performed once at each level at a particular direction, a total of 2 2 = 4 contourlets {C j i, j =,..., 4} at scale i will be produced (as shown in Figure 2), and their resolution will be a quarter of that of H i. The sample of derived four contoulets is shown in the lower right of Figure 2. It can be seen that, the size of these four contourlets is only a quarter of that of the input image I i, which is exactly the same as the ratio of the resolutions between adjacent image scales I i and I i. In our SR frameowrk, we perform contourlet transform for each image I i in Figure (a), and thus we construct the contourlets pyramid with four contourlets {C j i, i =, 2,... } for each scale, as shown in Figure (c). Together with our context information determined in Section II-A, we now obtain both context information and the associated contourlet for each image patch in the image pyramid (Figure (a)). For SR purposes, if we can properly predict the four directional components {C j 0, j =,..., 4} I i Bicubic Down-Sampling Low-Frequency Components High-Frequency Components U i I i- H i Bicubic Up-Sampling DFB... C i - 3 C i - 2 C i - 4 C i - Figure 2. A contourlet transform example decomposing I i into four contourlets {C j i, j =,..., 4} and the downsampled image I i (see Section II-B). from the LR input I 0, the high frequency component image H + can be synthesized by H + = 4 upsample(c j 0, Sj ) F j. (3) j= In contrast to the directional downsampling process in (2), the above equation reconstructs the image H + by convolving F j with the upsampled image of C j 0 using Sj. Note that F j are the synthesizing filter banks associated with E j and S j are the same operating matrices in (2). As a result, we can reconstruct the SR output using the recovered high-frequency image H + and the bicubic one U + (i.e., the up-sampled version of I 0 by bicubic interpolation). Therefore, the last step of our proposed method is to learn the relationship between the image patches of the same context category and their corresponding contourlet outputs. Once these models are observed, we can determine the context category for each image patch in I 0 (by the results of AP, as discussed in Section II-A), and apply the associated learning model to predict the contourlet. 2) Self-Learning of Context-Aware and Edge-Preserving SR Models: To learn the relationship between the image patches in I i of the same context category and their corresponding contourlet outputs C j i, we apply support vector regression (SVR) [] for this learning process. SVR has been observed an effective algorithm in fitting image data for SR purposes [6], [4]. For each context category given a LR input image, we consider their image patches in Figure (a) as the input data, and the corresponding conourlets in Figure (c) are the output labels to be observed. We extract the sparse representation of each image patch (of the same context category) as its feature vector, since sparse representation has been shown to produce promising results in image related applications especially SR [0]. To learn the sparse representation for an image patch, we solve the following optimization problem:

4 min 2 xg D g α g λ α g, g =, 2,..., k, (4) where x g is a 5 5 image patch (i.e., a 25 dimensional vector) of context label g (out of k categories), D g is the sparse coding dictionary to be learned, α g is the corresponding derived sparse coefficient vector, and the parameter λ balances the sparsity of α g. After the sparse representation features of image patches are calculated, the corresponding pixel values (at the same position of the center for the input patch) of the contourlets {C j i, j =,..., 4.} will be the label to be learned. We note that the pixel values of the contourlets indicate the intensity of the edge at a particular direction, which depicts the directional high frequency components for the image patches of a specific context category. We model the relationship between the sparse representation patches features and the contourlets by SVR, and thus we construct four SVR models for each context category. More precisely, we solve: n min w,b,ξ,ξ 2 wt w + C (ξ i + ξi ) (5) i= s.t. y g i (wt φ(α g i ) + b) ɛ + ξi, ξi 0, (w T φ(α g i ) + b) yg i ɛ + ξ i, ξi 0. In (5), y g i is the pixel values for the contourlet of context category g. The parameter b is the off-set of the regression model, n is the number of patches in context category g, and φ(α g i ) is the sparse representation of the input patch. The parameter w is the norm vector of the nonlinear mapping function, and C is the tradeoff between the generalization and upper/lower training errors ξ i /ξi with precision ɛ. Note that, we will have a total of 4k SVR models to learn given an input image (i.e., 4 contourlets k context categories). To accelerate the training process, we apply parallel processing techniques and significantly reduces the computation time, as shown in Figure (d). The procedure of recovering the SR output with the input I 0 is now described as follows. For each super-pixel in I 0, we first determine its context label using the AP clustering results on the pyramid I i (Figure (a)), and their sparse representation feature vector is calculated by (4). Next, we apply the corresponding SVR model to predict the contourlet for each patch, and four directional highfrequency components images {C j 0, j =,..., 4} will be obtained. These four contourlets will be used to reconstruct the high-frequency image H + with DFB reconstruction. As shown in Figure (e), this H + will be added back to the up-sampled image U + (by bicubic interpolation) to produce the final SR output. The advantage of learning the context-specific highfrequency components of the images without the intervention of the low-frequency ones is that we are able to achieve improved edge-preserving SR, as we verify later in Section III. Moreover, a self-learning framework like ours does not need to collect any training LR/HR image data beforehand (like most learning-based SR methods did), which makes our proposed method preferable in practical SR applications. III. EXPERIMENTAL RESULTS We apply our SR approach on images collected from USC-SIPI image database and Berkeley segmentation database [8]. The size of the test LR input image is pixels and the magnification factor is 2 (in one dimension). We note that the test images are obtained after applying Gaussian blur kernel on the original HR images ([2] also did this). For color images, only the illuminance channel in YIQ color space is used for SR, and the remaining color channels are up-sampled by bicubic interpolation. For learning sparse representation and SVR models (with Gaussian kernels), we apply the SPAMS package 2 and LIBSVM 3, respectively. The DFB decomposition and reconstruction are performed using the contourlet toolbox provided by Matlab 4. Moreover, for parallel processing implementation, we apply matlabpool to create a four-core environment for accelerating the computation time for our algorithm. Table I lists the PSNR values for a variety of images. To compare our approach with other existing interpolation or learning-based SR methods, we consider bicubic interpolation, locally linear embedding (LLE) for SR [4], learning image sparse representation for SR [9], a learning-based single image SR approach [4], and the method of Glasner et al. [2]. From this table, it is clear that our method achieved the best PSNR values for all images except for airplane. Compared to bicubic interpolation, we obtained an average PSNR improvement of 2.9% (or about db), which is remarkably better than other learning-based SR methods. We show example SR results and comparisons in Figures 3, 4, and 5. From these figures, we see that our approach produced HR images with clear edges and less artifacts that other methods did, and thus qualitatively better SR results were achieved. We observe similar remarks applying our method to synthesize higher-resolution images (with a magnification factor > 2). Furthermore, our proposed method achieved a very impressive SR efficiency compared with other learningbased approaches. Our SR algorithm reported an average processing (testing) time of 270 seconds (without parallel processing), which is significantly less than the other two single image learning-based methods [4] and [2]. For other learning-based approaches requiring training data (and thus the training time needs to be considered), ours is still much more computationally efficient ( seconds for [4] and over 0 5 seconds for [9]). It is worth repeating that, even these two methods only need to perform training Available at 2 Software available at 3 C.-C. Chang and C.-J. Lin, LIBSVM: a library for SVMs,

5 Table I PSNR AND COMPUTATION TIME COMPARISONS OF DIFFERENT SR METHODS. NOTE THAT BOTH THE AVERAGE TRAINING AND TESTING TIME (T T rain AND T T est ) ARE LISTED FOR EACH METHOD (IF APPLICABLE), AND THE RUN-TIME ESTIMATES (IN SECONDS) WERE OBTAINED ON AN INTEL QUAD CORE PC WITH 3 GHZ CPU AND 4G RAM. Training boat cars skyview lena person airplane child T T rain T T est Image Bicubic N/A N/A N/A LLE [4] Yes Yang et al. [9] Yes Wang et al. [4] N/A 4800 No Glasner et al. [2] N/A 5400 No Our method (without parallel processing) N/A 270 No Our method (with parallel processing) once, their performance will be sensitive to their collection of training data and thus might not be practical for realworld SR applications. Finally, once we integrate the parallel processing technique into our proposed SR framework, we only require about 80 seconds to synthesize a HR output using a single LR input image. From the above qualitative and quantitative comparisons, we can verify the effectiveness and efficiency of our proposed SR method. IV. CONCLUSION We proposed a self-learning approach for single image super-resolution with contourlet transform. By utilizing the contourlets to model directional high-frequency components present in an input LR image (and its image pyramid), our SR framework is able to learn context-specific highfrequency details of images patches extracted from different context categories. Using sparse representation as image features and support vector regression as the learning models, our proposed method models the relationship between image patches and the associated contourlets for each context category. These observed models are applied to predict the directional high-frequency components for each patch in the input LR image, and the results are used to synthesize/refine the final SR output. Our SR approach is very unique and different from conventional learningbased approaches, since we do not assume any image priors such as the reoccurrence/self-similarity of image patches, nor do we need to collect training LR/HR image data in advance. Our experimental results verified the effectiveness and efficiency of our method, and we showed that our approach qualitatively and quantitatively outperforms stateof-the-art learning-based SR methods. REFERENCES [] R. C. Hardie et al., Joint map registration and high-resolution image estimation using a sequence of undersampled images, IEEE Trans. Image Processing, 997. [2] S. Farsiu et al., Fast and robust multi-frame super-resolution, IEEE Trans. Image Processing, [3] W. T. Freeman, T. Jones, and E. Pasztor, Example-based super-resolution, IEEE Computer Graphics & Applications, [4] H. Chang, D.-Y. Yeung, and Y. Xiong, Super-resolution through neighbor embedding, in IEEE CVPR, [5] C. V. Jiji and S. Chaudhuri, Single-Frame Image Superresolution through Contourlet Learning, EURASIP J. Adv. Sig. Proc., [6] K. Ni and T. Q. Nguyen, Image superresolution using support vector regression, IEEE Trans. Image Processing, [7] K. P. Upla, P. P. Gajjar, M. V. Joshi, A. Banerjee and V. Singh, A fast approach for edge preserving super-resolution, in IEEE ICME, 20. [8] M. N. Do and M. Vetterli, The contourlet transform: an efficient directional multiresolution image representation, IEEE Trans. Image Processing, [9] J. Yang, J. Wright, T. Huang, and Y. Ma, Image superresolution via sparse representation, IEEE Trans. Image Processing, 200. [0] M.-C. Yang, C.-T. Chu and Y.-C. F. Wang, Learning sparse image representation with support vector regression for singleimage super-resolution, in IEEE ICIP, 200. [] V. Vapnik, Statistical Learning Theory, Wiley-Interscience, 998. [2] D. Glasner, S. Bagon, and M. Irani, Super-resolution from a single image, in IEEE ICCV, [3] C.-Y. Yang, J.-B. Huang and M.-H. Yang, Exploiting selfsimilarities for single frame super-resolution, in Asian Conf. Computer Vision, 200. [4] M.-C. Yang, C.-H. Wang, T.-Y. Hu and Y.-C. F. Wang, Learning Context-Aware Sparse Representation for Single Image Super-Resolution, in IEEE ICIP, 20. [5] J. Sun et al., Context-constrained hallucination for image super-resolution, in IEEE CVPR, 200. [6] D. Comaniciu and P. Meer, Mean shift: A robust approach toward feature space analysis, IEEE Trans. Pattern Analysis Machine Intelligence, [7] D. Dueck and B. J. Frey, Clustering by passing messages between data points, Science, [8] D. Martin et al., A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics, in IEEE ICCV, 200.

6 Figure 3. Example SR results (with a magnification factor of 2) and the corresponding PSNR values. Images from left to right: Ground truth, Yang et al. [9] (PSNR: 25.32), Glasner et al. [2] (PSNR: 27.64) and ours (PSNR: 27.92). Note that the method of Yang et al. [9] requires the training LR/HR image data, while Glasner et al. [2] and our approach are learning-based methods for single image super-resolution (i.e., no training LR/HR image data is needed). Figure 4. Example SR results (with a magnification factor 2) and the corresponding PSNR values. Images from left to right: Ground truth, Yang et al. [9] (PSNR: 28.86), Glasner et al. [2] (PSNR: 34.98) and ours (PSNR: 35.6). Figure 5. Example SR results (with a magnification factor of 2) and the corresponding PSNR values. Images from left to right: Ground truth, Yang et al. [9] (PSNR: 3.28), Glasner et al. [2] (PSNR: 25.54) and ours (PSNR: 36.33).

A Self-Learning Optimization Approach to Single Image Super-Resolution using Kernel ridge regression model

A Self-Learning Optimization Approach to Single Image Super-Resolution using Kernel ridge regression model A Self-Learning Optimization Approach to Single Image Super-Resolution using Kernel ridge regression model Ms. Dharani S 1 PG Student (CSE), Sri Krishna College of Engineering and Technology, Anna University,

More information

Ms.DHARANI SAMPATH Computer Science And Engineering, Sri Krishna College Of Engineering & Technology Coimbatore, India

Ms.DHARANI SAMPATH Computer Science And Engineering, Sri Krishna College Of Engineering & Technology Coimbatore, India Improving Super Resolution of Image by Multiple Kernel Learning Ms.DHARANI SAMPATH Computer Science And Engineering, Sri Krishna College Of Engineering & Technology Coimbatore, India dharanis012@gmail.com

More information

Learning based face hallucination techniques: A survey

Learning based face hallucination techniques: A survey Vol. 3 (2014-15) pp. 37-45. : A survey Premitha Premnath K Department of Computer Science & Engineering Vidya Academy of Science & Technology Thrissur - 680501, Kerala, India (email: premithakpnath@gmail.com)

More information

Example-Based Image Super-Resolution Techniques

Example-Based Image Super-Resolution Techniques Example-Based Image Super-Resolution Techniques Mark Sabini msabini & Gili Rusak gili December 17, 2016 1 Introduction With the current surge in popularity of imagebased applications, improving content

More information

Single-Image Super-Resolution Using Multihypothesis Prediction

Single-Image Super-Resolution Using Multihypothesis Prediction Single-Image Super-Resolution Using Multihypothesis Prediction Chen Chen and James E. Fowler Department of Electrical and Computer Engineering, Geosystems Research Institute (GRI) Mississippi State University,

More information

Enhancing DubaiSat-1 Satellite Imagery Using a Single Image Super-Resolution

Enhancing DubaiSat-1 Satellite Imagery Using a Single Image Super-Resolution Enhancing DubaiSat-1 Satellite Imagery Using a Single Image Super-Resolution Saeed AL-Mansoori 1 and Alavi Kunhu 2 1 Associate Image Processing Engineer, SIPAD Image Enhancement Section Emirates Institution

More information

FAST SINGLE-IMAGE SUPER-RESOLUTION WITH FILTER SELECTION. Image Processing Lab Technicolor R&I Hannover

FAST SINGLE-IMAGE SUPER-RESOLUTION WITH FILTER SELECTION. Image Processing Lab Technicolor R&I Hannover FAST SINGLE-IMAGE SUPER-RESOLUTION WITH FILTER SELECTION Jordi Salvador Eduardo Pérez-Pellitero Axel Kochale Image Processing Lab Technicolor R&I Hannover ABSTRACT This paper presents a new method for

More information

Exploiting Self-Similarities for Single Frame Super-Resolution

Exploiting Self-Similarities for Single Frame Super-Resolution Exploiting Self-Similarities for Single Frame Super-Resolution Chih-Yuan Yang Jia-Bin Huang Ming-Hsuan Yang Electrical Engineering and Computer Science University of California at Merced Merced, CA 95343,

More information

Structured Face Hallucination

Structured Face Hallucination 2013 IEEE Conference on Computer Vision and Pattern Recognition Structured Face Hallucination Chih-Yuan Yang Sifei Liu Ming-Hsuan Yang Electrical Engineering and Computer Science University of California

More information

Interpolation Based Image Super Resolution by Support-Vector-Regression

Interpolation Based Image Super Resolution by Support-Vector-Regression Interpolation Based Image Super Resolution by Support-Vector-Regression Sowmya. M 1, Anand M.J 2 1 Final Year M.Tech Student, Department Of Electronics And Communication, PES College Of Engineering, Mandya,

More information

Augmented Coupled Dictionary Learning for Image Super-Resolution

Augmented Coupled Dictionary Learning for Image Super-Resolution Augmented Coupled Dictionary Learning for Image Super-Resolution Muhammad Rushdi and Jeffrey Ho Computer and Information Science and Engineering University of Florida Gainesville, Florida, U.S.A. Email:

More information

Anchored Neighborhood Regression for Fast Example-Based Super-Resolution

Anchored Neighborhood Regression for Fast Example-Based Super-Resolution Anchored Neighborhood Regression for Fast Example-Based Super-Resolution Radu Timofte 1,2, Vincent De Smet 1, and Luc Van Gool 1,2 1 KU Leuven, ESAT-PSI / iminds, VISICS 2 ETH Zurich, D-ITET, Computer

More information

IMAGE SUPER-RESOLUTION BASED ON DICTIONARY LEARNING AND ANCHORED NEIGHBORHOOD REGRESSION WITH MUTUAL INCOHERENCE

IMAGE SUPER-RESOLUTION BASED ON DICTIONARY LEARNING AND ANCHORED NEIGHBORHOOD REGRESSION WITH MUTUAL INCOHERENCE IMAGE SUPER-RESOLUTION BASED ON DICTIONARY LEARNING AND ANCHORED NEIGHBORHOOD REGRESSION WITH MUTUAL INCOHERENCE Yulun Zhang 1, Kaiyu Gu 2, Yongbing Zhang 1, Jian Zhang 3, and Qionghai Dai 1,4 1 Shenzhen

More information

Super-resolution using Neighbor Embedding of Back-projection residuals

Super-resolution using Neighbor Embedding of Back-projection residuals Super-resolution using Neighbor Embedding of Back-projection residuals Marco Bevilacqua, Aline Roumy, Christine Guillemot SIROCCO Research team INRIA Rennes, France {marco.bevilacqua, aline.roumy, christine.guillemot}@inria.fr

More information

arxiv: v1 [cs.cv] 8 Feb 2018

arxiv: v1 [cs.cv] 8 Feb 2018 DEEP IMAGE SUPER RESOLUTION VIA NATURAL IMAGE PRIORS Hojjat S. Mousavi, Tiantong Guo, Vishal Monga Dept. of Electrical Engineering, The Pennsylvania State University arxiv:802.0272v [cs.cv] 8 Feb 208 ABSTRACT

More information

Robust Single Image Super-resolution based on Gradient Enhancement

Robust Single Image Super-resolution based on Gradient Enhancement Robust Single Image Super-resolution based on Gradient Enhancement Licheng Yu, Hongteng Xu, Yi Xu and Xiaokang Yang Department of Electronic Engineering, Shanghai Jiaotong University, Shanghai 200240,

More information

Single Image Super-Resolution via Sparse Representation over Directionally Structured Dictionaries Based on the Patch Gradient Phase Angle

Single Image Super-Resolution via Sparse Representation over Directionally Structured Dictionaries Based on the Patch Gradient Phase Angle 2014 UKSim-AMSS 8th European Modelling Symposium Single Image Super-Resolution via Sparse Representation over Directionally Structured Dictionaries Based on the Patch Gradient Phase Angle Mahmoud Nazzal,

More information

Image Interpolation Using Multiscale Geometric Representations

Image Interpolation Using Multiscale Geometric Representations Image Interpolation Using Multiscale Geometric Representations Nickolaus Mueller, Yue Lu and Minh N. Do Department of Electrical and Computer Engineering University of Illinois at Urbana-Champaign ABSTRACT

More information

Image Interpolation using Collaborative Filtering

Image Interpolation using Collaborative Filtering Image Interpolation using Collaborative Filtering 1,2 Qiang Guo, 1,2,3 Caiming Zhang *1 School of Computer Science and Technology, Shandong Economic University, Jinan, 250014, China, qguo2010@gmail.com

More information

Comparative Analysis of Edge Based Single Image Superresolution

Comparative Analysis of Edge Based Single Image Superresolution Comparative Analysis of Edge Based Single Image Superresolution Sonali Shejwal 1, Prof. A. M. Deshpande 2 1,2 Department of E&Tc, TSSM s BSCOER, Narhe, University of Pune, India. ABSTRACT: Super-resolution

More information

A Novel Multi-Frame Color Images Super-Resolution Framework based on Deep Convolutional Neural Network. Zhe Li, Shu Li, Jianmin Wang and Hongyang Wang

A Novel Multi-Frame Color Images Super-Resolution Framework based on Deep Convolutional Neural Network. Zhe Li, Shu Li, Jianmin Wang and Hongyang Wang 5th International Conference on Measurement, Instrumentation and Automation (ICMIA 2016) A Novel Multi-Frame Color Images Super-Resolution Framewor based on Deep Convolutional Neural Networ Zhe Li, Shu

More information

Single Image Super-resolution. Slides from Libin Geoffrey Sun and James Hays

Single Image Super-resolution. Slides from Libin Geoffrey Sun and James Hays Single Image Super-resolution Slides from Libin Geoffrey Sun and James Hays Cs129 Computational Photography James Hays, Brown, fall 2012 Types of Super-resolution Multi-image (sub-pixel registration) Single-image

More information

ROBUST INTERNAL EXEMPLAR-BASED IMAGE ENHANCEMENT. Yang Xian 1 and Yingli Tian 1,2

ROBUST INTERNAL EXEMPLAR-BASED IMAGE ENHANCEMENT. Yang Xian 1 and Yingli Tian 1,2 ROBUST INTERNAL EXEMPLAR-BASED IMAGE ENHANCEMENT Yang Xian 1 and Yingli Tian 1,2 1 The Graduate Center, 2 The City College, The City University of New York, New York, Email: yxian@gc.cuny.edu; ytian@ccny.cuny.edu

More information

Boosting face recognition via neural Super-Resolution

Boosting face recognition via neural Super-Resolution Boosting face recognition via neural Super-Resolution Guillaume Berger, Cle ment Peyrard and Moez Baccouche Orange Labs - 4 rue du Clos Courtel, 35510 Cesson-Se vigne - France Abstract. We propose a two-step

More information

Bilevel Sparse Coding

Bilevel Sparse Coding Adobe Research 345 Park Ave, San Jose, CA Mar 15, 2013 Outline 1 2 The learning model The learning algorithm 3 4 Sparse Modeling Many types of sensory data, e.g., images and audio, are in high-dimensional

More information

Texture Sensitive Image Inpainting after Object Morphing

Texture Sensitive Image Inpainting after Object Morphing Texture Sensitive Image Inpainting after Object Morphing Yin Chieh Liu and Yi-Leh Wu Department of Computer Science and Information Engineering National Taiwan University of Science and Technology, Taiwan

More information

MULTI-POSE FACE HALLUCINATION VIA NEIGHBOR EMBEDDING FOR FACIAL COMPONENTS. Yanghao Li, Jiaying Liu, Wenhan Yang, Zongming Guo

MULTI-POSE FACE HALLUCINATION VIA NEIGHBOR EMBEDDING FOR FACIAL COMPONENTS. Yanghao Li, Jiaying Liu, Wenhan Yang, Zongming Guo MULTI-POSE FACE HALLUCINATION VIA NEIGHBOR EMBEDDING FOR FACIAL COMPONENTS Yanghao Li, Jiaying Liu, Wenhan Yang, Zongg Guo Institute of Computer Science and Technology, Peking University, Beijing, P.R.China,

More information

Image Super-Resolution via Sparse Representation

Image Super-Resolution via Sparse Representation Image Super-Resolution via Sparse Representation Jianchao Yang, John Wright, Thomas Huang and Yi Ma accepted by IEEE Trans. on Image Processing 2010 Presented by known 2010/4/20 1 Super-Resolution Techniques

More information

Improved Feature Selection for Neighbor Embedding Super-Resolution Using Zernike Moments

Improved Feature Selection for Neighbor Embedding Super-Resolution Using Zernike Moments Improved Feature Selection for Neighbor Embedding Super-Resolution Using Zernike Moments Deepasikha Mishra, Banshidhar Majhi and Pankaj Kumar Sa Abstract This paper presents a new feature selection method

More information

Single Image Super Resolution with Wavelet Domain Transformation and Sparse Representation

Single Image Super Resolution with Wavelet Domain Transformation and Sparse Representation International Journal of Innovative Research in Computer Science & Technology (IJIRCST) ISSN: 2347-5552, Volume-4, Issue-1, January 2016 Single Image Super Resolution with Wavelet Domain Transformation

More information

Fast Image Super-resolution Based on In-place Example Regression

Fast Image Super-resolution Based on In-place Example Regression 2013 IEEE Conference on Computer Vision and Pattern Recognition Fast Image Super-resolution Based on In-place Example Regression Jianchao Yang, Zhe Lin, Scott Cohen Adobe Research 345 Park Avenue, San

More information

Speed up a Machine-Learning-based Image Super-Resolution Algorithm on GPGPU

Speed up a Machine-Learning-based Image Super-Resolution Algorithm on GPGPU Speed up a Machine-Learning-based Image Super-Resolution Algorithm on GPGPU Ke Ma 1, and Yao Song 2 1 Department of Computer Sciences 2 Department of Electrical and Computer Engineering University of Wisconsin-Madison

More information

Super-resolution using Neighbor Embedding of Back-projection residuals

Super-resolution using Neighbor Embedding of Back-projection residuals Super-resolution using Neighbor Embedding of Back-projection residuals Marco Bevilacqua, Aline Roumy, Christine Guillemot, Marie-Line Alberi Morel To cite this version: Marco Bevilacqua, Aline Roumy, Christine

More information

Single Image Super-Resolution

Single Image Super-Resolution Single Image Super-Resolution Abhishek Arora Dept. of Electrical Engg. Stanford University, CA Email: arorabhi@stanford.edu Ritesh Kolte Dept. of Electrical Engg. Stanford University, CA Email: rkolte@stanford.edu

More information

Sparse Representation Based Super-Resolution Algorithm using Wavelet Domain Interpolation and Nonlocal Means

Sparse Representation Based Super-Resolution Algorithm using Wavelet Domain Interpolation and Nonlocal Means TELKOMNIKA Indonesian Journal of Electrical Engineering Vol. 16, No. 2, November 2015, pp. 296 ~ 302 DOI: 10.11591/telkomnika.v16i2.8816 296 Sparse Representation Based Super-Resolution Algorithm using

More information

Influence of Training Set and Iterative Back Projection on Example-based Super-resolution

Influence of Training Set and Iterative Back Projection on Example-based Super-resolution Influence of Training Set and Iterative Back Projection on Example-based Super-resolution Saeid Fazli Research Institute of Modern Biological Techniques University of zanjan Zanjan, Iran Abstract Example-based

More information

Single Image Super-resolution using Deformable Patches

Single Image Super-resolution using Deformable Patches Single Image Super-resolution using Deformable Patches Yu Zhu 1, Yanning Zhang 1, Alan L. Yuille 2 1 School of Computer Science, Northwestern Polytechnical University, China 2 Department of Statistics,

More information

Super-Resolution from a Single Image

Super-Resolution from a Single Image Super-Resolution from a Single Image Daniel Glasner Shai Bagon Michal Irani Dept. of Computer Science and Applied Mathematics The Weizmann Institute of Science Rehovot 76100, Israel Abstract Methods for

More information

CS 534: Computer Vision Texture

CS 534: Computer Vision Texture CS 534: Computer Vision Texture Ahmed Elgammal Dept of Computer Science CS 534 Texture - 1 Outlines Finding templates by convolution What is Texture Co-occurrence matrices for texture Spatial Filtering

More information

Guided Image Super-Resolution: A New Technique for Photogeometric Super-Resolution in Hybrid 3-D Range Imaging

Guided Image Super-Resolution: A New Technique for Photogeometric Super-Resolution in Hybrid 3-D Range Imaging Guided Image Super-Resolution: A New Technique for Photogeometric Super-Resolution in Hybrid 3-D Range Imaging Florin C. Ghesu 1, Thomas Köhler 1,2, Sven Haase 1, Joachim Hornegger 1,2 04.09.2014 1 Pattern

More information

Learning and Inferring Depth from Monocular Images. Jiyan Pan April 1, 2009

Learning and Inferring Depth from Monocular Images. Jiyan Pan April 1, 2009 Learning and Inferring Depth from Monocular Images Jiyan Pan April 1, 2009 Traditional ways of inferring depth Binocular disparity Structure from motion Defocus Given a single monocular image, how to infer

More information

A Bayesian Approach to Alignment-based Image Hallucination

A Bayesian Approach to Alignment-based Image Hallucination A Bayesian Approach to Alignment-based Image Hallucination Marshall F. Tappen 1 and Ce Liu 2 1 University of Central Florida mtappen@eecs.ucf.edu 2 Microsoft Research New England celiu@microsoft.com Abstract.

More information

A A A. Fig.1 image patch. Then the edge gradient magnitude is . (1)

A A A. Fig.1 image patch. Then the edge gradient magnitude is . (1) International Conference on Information Science and Computer Applications (ISCA 013) Two-Dimensional Barcode Image Super-Resolution Reconstruction Via Sparse Representation Gaosheng Yang 1,Ningzhong Liu

More information

COMPRESSED FACE HALLUCINATION. Electrical Engineering and Computer Science University of California, Merced, CA 95344, USA

COMPRESSED FACE HALLUCINATION. Electrical Engineering and Computer Science University of California, Merced, CA 95344, USA COMPRESSED FACE HALLUCNATON Sifei Liu Ming-Hsuan Yang Electrical Engineering and Computer Science University of California, Merced, CA 95344, USA ABSTRACT n this paper, we propose an algorithm to hallucinate

More information

NTHU Rain Removal Project

NTHU Rain Removal Project People NTHU Rain Removal Project Networked Video Lab, National Tsing Hua University, Hsinchu, Taiwan Li-Wei Kang, Institute of Information Science, Academia Sinica, Taipei, Taiwan Chia-Wen Lin *, Department

More information

IMAGE ENHANCEMENT USING NONSUBSAMPLED CONTOURLET TRANSFORM

IMAGE ENHANCEMENT USING NONSUBSAMPLED CONTOURLET TRANSFORM IMAGE ENHANCEMENT USING NONSUBSAMPLED CONTOURLET TRANSFORM Rafia Mumtaz 1, Raja Iqbal 2 and Dr.Shoab A.Khan 3 1,2 MCS, National Unioversity of Sciences and Technology, Rawalpindi, Pakistan: 3 EME, National

More information

Super-Resolution. Many slides from Miki Elad Technion Yosi Rubner RTC and more

Super-Resolution. Many slides from Miki Elad Technion Yosi Rubner RTC and more Super-Resolution Many slides from Mii Elad Technion Yosi Rubner RTC and more 1 Example - Video 53 images, ratio 1:4 2 Example Surveillance 40 images ratio 1:4 3 Example Enhance Mosaics 4 5 Super-Resolution

More information

INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET)

INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET) INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET) ISSN 0976 6464(Print) ISSN 0976 6472(Online) Volume 3, Issue 3, October- December (2012), pp. 153-161 IAEME: www.iaeme.com/ijecet.asp

More information

Image Deblurring Using Adaptive Sparse Domain Selection and Adaptive Regularization

Image Deblurring Using Adaptive Sparse Domain Selection and Adaptive Regularization Volume 3, No. 3, May-June 2012 International Journal of Advanced Research in Computer Science RESEARCH PAPER Available Online at www.ijarcs.info ISSN No. 0976-5697 Image Deblurring Using Adaptive Sparse

More information

arxiv: v1 [cs.cv] 3 Jan 2017

arxiv: v1 [cs.cv] 3 Jan 2017 Learning a Mixture of Deep Networks for Single Image Super-Resolution Ding Liu, Zhaowen Wang, Nasser Nasrabadi, and Thomas Huang arxiv:1701.00823v1 [cs.cv] 3 Jan 2017 Beckman Institute, University of Illinois

More information

An efficient face recognition algorithm based on multi-kernel regularization learning

An efficient face recognition algorithm based on multi-kernel regularization learning Acta Technica 61, No. 4A/2016, 75 84 c 2017 Institute of Thermomechanics CAS, v.v.i. An efficient face recognition algorithm based on multi-kernel regularization learning Bi Rongrong 1 Abstract. A novel

More information

AUTOMATIC OBJECT EXTRACTION IN SINGLE-CONCEPT VIDEOS. Kuo-Chin Lien and Yu-Chiang Frank Wang

AUTOMATIC OBJECT EXTRACTION IN SINGLE-CONCEPT VIDEOS. Kuo-Chin Lien and Yu-Chiang Frank Wang AUTOMATIC OBJECT EXTRACTION IN SINGLE-CONCEPT VIDEOS Kuo-Chin Lien and Yu-Chiang Frank Wang Research Center for Information Technology Innovation, Academia Sinica, Taipei, Taiwan {iker, ycwang}@citi.sinica.edu.tw

More information

IMAGE SUPER RESOLUTION USING NON SUB-SAMPLE CONTOURLET TRANSFORM WITH LOCAL TERNARY PATTERN

IMAGE SUPER RESOLUTION USING NON SUB-SAMPLE CONTOURLET TRANSFORM WITH LOCAL TERNARY PATTERN IMAGE SUPER RESOLUTION USING NON SUB-SAMPLE CONTOURLET TRANSFORM WITH LOCAL TERNARY PATTERN Pikin S. Patel 1, Parul V. Pithadia 2, Manoj parmar 3 PG. Student, EC Dept., Dr. S & S S Ghandhy Govt. Engg.

More information

Image Super-Resolution Reconstruction Based On L 1/2 Sparsity

Image Super-Resolution Reconstruction Based On L 1/2 Sparsity Buletin Teknik Elektro dan Informatika (Bulletin of Electrical Engineering and Informatics) Vol. 3, No. 3, September 4, pp. 55~6 ISSN: 89-39 55 Image Super-Resolution Reconstruction Based On L / Sparsity

More information

Text Super-Resolution and Deblurring using Multiple Support Vector Regression

Text Super-Resolution and Deblurring using Multiple Support Vector Regression Text Super-Resolution and Deblurring using Multiple Support Vector Regression Roy Blankman Sean McMillan Ross Smith December 15 th, 2011 1 Introduction In a world that is increasingly digital, there is

More information

arxiv: v1 [cs.cv] 6 Nov 2015

arxiv: v1 [cs.cv] 6 Nov 2015 Seven ways to improve example-based single image super resolution Radu Timofte Computer Vision Lab D-ITET, ETH Zurich timofter@vision.ee.ethz.ch Rasmus Rothe Computer Vision Lab D-ITET, ETH Zurich rrothe@vision.ee.ethz.ch

More information

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 24, NO. 9, SEPTEMBER

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 24, NO. 9, SEPTEMBER IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 24, NO. 9, SEPTEMBER 2015 2797 Image Super-Resolution Based on Structure-Modulated Sparse Representation Yongqin Zhang, Member, IEEE, Jiaying Liu, Member, IEEE,

More information

An Approach for Reduction of Rain Streaks from a Single Image

An Approach for Reduction of Rain Streaks from a Single Image An Approach for Reduction of Rain Streaks from a Single Image Vijayakumar Majjagi 1, Netravati U M 2 1 4 th Semester, M. Tech, Digital Electronics, Department of Electronics and Communication G M Institute

More information

Single Image Improvement using Superresolution.

Single Image Improvement using Superresolution. Single Image Improvement using Superresolution. ABSTRACT Shwetambari Shinde, Meeta Dewangan Department of Computer Science & Engineering,CSIT,Bhilai,India. shweta_shinde9388@yahoo Department of Computer

More information

Single image super-resolution by directionally structured coupled dictionary learning

Single image super-resolution by directionally structured coupled dictionary learning Ahmed and Shah EURASIP Journal on Image and Video Processing (2016) 2016:36 DOI 10.1186/s13640-016-0141-6 EURASIP Journal on Image and Video Processing RESEARCH Open Access Single image super-resolution

More information

Super-resolution via Transform-invariant Group-sparse Regularization

Super-resolution via Transform-invariant Group-sparse Regularization 2013 IEEE International Conference on Computer Vision Super-resolution via Transform-invariant Group-sparse Regularization Carlos Fernandez-Granda Stanford University cfgranda@stanford.edu Emmanuel J.

More information

A Bayesian Approach to Alignment-Based Image Hallucination

A Bayesian Approach to Alignment-Based Image Hallucination A Bayesian Approach to Alignment-Based Image Hallucination Marshall F. Tappen 1 and Ce Liu 2 1 University of Central Florida mtappen@eecs.ucf.edu 2 Microsoft Research New England celiu@microsoft.com Abstract.

More information

Storage Efficient NL-Means Burst Denoising for Programmable Cameras

Storage Efficient NL-Means Burst Denoising for Programmable Cameras Storage Efficient NL-Means Burst Denoising for Programmable Cameras Brendan Duncan Stanford University brendand@stanford.edu Miroslav Kukla Stanford University mkukla@stanford.edu Abstract An effective

More information

Anisotropic representations for superresolution of hyperspectral data

Anisotropic representations for superresolution of hyperspectral data Anisotropic representations for superresolution of hyperspectral data Edward H. Bosch, Wojciech Czaja, James M. Murphy, and Daniel Weinberg Norbert Wiener Center Department of Mathematics University of

More information

Single-image super-resolution in RGB space via group sparse representation

Single-image super-resolution in RGB space via group sparse representation Published in IET Image Processing Received on 10th April 2014 Revised on 22nd September 2014 Accepted on 15th October 2014 ISSN 1751-9659 Single-image super-resolution in RGB space via group sparse representation

More information

Image Inpainting by Hyperbolic Selection of Pixels for Two Dimensional Bicubic Interpolations

Image Inpainting by Hyperbolic Selection of Pixels for Two Dimensional Bicubic Interpolations Image Inpainting by Hyperbolic Selection of Pixels for Two Dimensional Bicubic Interpolations Mehran Motmaen motmaen73@gmail.com Majid Mohrekesh mmohrekesh@yahoo.com Mojtaba Akbari mojtaba.akbari@ec.iut.ac.ir

More information

Super Resolution Using Graph-cut

Super Resolution Using Graph-cut Super Resolution Using Graph-cut Uma Mudenagudi, Ram Singla, Prem Kalra, and Subhashis Banerjee Department of Computer Science and Engineering Indian Institute of Technology Delhi Hauz Khas, New Delhi,

More information

Fast Learning-Based Single Image Super-Resolution

Fast Learning-Based Single Image Super-Resolution 1 Fast Learning-Based Single Image Super-Resolution Neeraj Kumar and Amit Sethi Abstract We present a learning-based single image superresolution (SISR) method to obtain a high resolution (HR) image from

More information

Image Resizing Based on Gradient Vector Flow Analysis

Image Resizing Based on Gradient Vector Flow Analysis Image Resizing Based on Gradient Vector Flow Analysis Sebastiano Battiato battiato@dmi.unict.it Giovanni Puglisi puglisi@dmi.unict.it Giovanni Maria Farinella gfarinellao@dmi.unict.it Daniele Ravì rav@dmi.unict.it

More information

IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING

IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING Idan Ram, Michael Elad and Israel Cohen Department of Electrical Engineering Department of Computer Science Technion - Israel Institute of Technology

More information

A Single Image Compression Framework Combined with Sparse Representation-Based Super- Resolution

A Single Image Compression Framework Combined with Sparse Representation-Based Super- Resolution International Conference on Electronic Science and Automation Control (ESAC 2015) A Single Compression Framework Combined with Sparse RepresentationBased Super Resolution He Xiaohai, He Jingbo, Huang Jianqiu

More information

K-Means Based Matching Algorithm for Multi-Resolution Feature Descriptors

K-Means Based Matching Algorithm for Multi-Resolution Feature Descriptors K-Means Based Matching Algorithm for Multi-Resolution Feature Descriptors Shao-Tzu Huang, Chen-Chien Hsu, Wei-Yen Wang International Science Index, Electrical and Computer Engineering waset.org/publication/0007607

More information

Efficient Module Based Single Image Super Resolution for Multiple Problems

Efficient Module Based Single Image Super Resolution for Multiple Problems Efficient Module Based Single Image Super Resolution for Multiple Problems Dongwon Park Kwanyoung Kim Se Young Chun School of ECE, Ulsan National Institute of Science and Technology, 44919, Ulsan, South

More information

A Novel Image Super-resolution Reconstruction Algorithm based on Modified Sparse Representation

A Novel Image Super-resolution Reconstruction Algorithm based on Modified Sparse Representation , pp.162-167 http://dx.doi.org/10.14257/astl.2016.138.33 A Novel Image Super-resolution Reconstruction Algorithm based on Modified Sparse Representation Liqiang Hu, Chaofeng He Shijiazhuang Tiedao University,

More information

IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS

IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS P.Mahalakshmi 1, J.Muthulakshmi 2, S.Kannadhasan 3 1,2 U.G Student, 3 Assistant Professor, Department of Electronics

More information

Seven ways to improve example-based single image super resolution

Seven ways to improve example-based single image super resolution Seven ways to improve example-based single image super resolution Radu Timofte CVL, D-ITET, ETH Zurich radu.timofte@vision.ee.ethz.ch Rasmus Rothe CVL, D-ITET, ETH Zurich rrothe@vision.ee.ethz.ch Luc Van

More information

Single Image Super-Resolution via Iterative Collaborative Representation

Single Image Super-Resolution via Iterative Collaborative Representation Single Image Super-Resolution via Iterative Collaborative Representation Yulun Zhang 1(B), Yongbing Zhang 1, Jian Zhang 2, aoqian Wang 1, and Qionghai Dai 1,3 1 Graduate School at Shenzhen, Tsinghua University,

More information

Image Super-Resolution using Gradient Profile Prior

Image Super-Resolution using Gradient Profile Prior Image Super-Resolution using Gradient Profile Prior Jian Sun 1 Jian Sun 2 Zongben Xu 1 Heung-Yeung Shum 2 1 Xi an Jiaotong University 2 Microsoft Research Asia Xi an, P. R. China Beijing, P. R. China Abstract

More information

arxiv: v2 [cs.cv] 11 Nov 2016

arxiv: v2 [cs.cv] 11 Nov 2016 Accurate Image Super-Resolution Using Very Deep Convolutional Networks Jiwon Kim, Jung Kwon Lee and Kyoung Mu Lee Department of ECE, ASRI, Seoul National University, Korea {j.kim, deruci, kyoungmu}@snu.ac.kr

More information

On Single Image Scale-Up using Sparse-Representation

On Single Image Scale-Up using Sparse-Representation On Single Image Scale-Up using Sparse-Representation Roman Zeyde, Matan Protter and Michael Elad The Computer Science Department Technion Israel Institute of Technology Haifa 32000, Israel {romanz,matanpr,elad}@cs.technion.ac.il

More information

Patch Based Blind Image Super Resolution

Patch Based Blind Image Super Resolution Patch Based Blind Image Super Resolution Qiang Wang, Xiaoou Tang, Harry Shum Microsoft Research Asia, Beijing 100080, P.R. China {qiangwa,xitang,hshum@microsoft.com} Abstract In this paper, a novel method

More information

Improved Super-Resolution through Residual Neighbor Embedding

Improved Super-Resolution through Residual Neighbor Embedding Improved Super-Resolution through Residual Neighbor Embedding Tak-Ming Chan 1 and Junping Zhang 1 2 1 Shanghai Key Laboratory of Intelligent Information Processing, Department of Computer Science and Engineering,

More information

COMPACT AND COHERENT DICTIONARY CONSTRUCTION FOR EXAMPLE-BASED SUPER-RESOLUTION

COMPACT AND COHERENT DICTIONARY CONSTRUCTION FOR EXAMPLE-BASED SUPER-RESOLUTION COMPACT AND COHERENT DICTIONARY CONSTRUCTION FOR EXAMPLE-BASED SUPER-RESOLUTION Marco Bevilacqua Aline Roumy Christine Guillemot Marie-Line Alberi Morel INRIA Rennes, Campus de Beaulieu, 35042 Rennes Cedex,

More information

Resolution Magnification Technique for Satellite Images Using DT- CWT and NLM

Resolution Magnification Technique for Satellite Images Using DT- CWT and NLM AUSTRALIAN JOURNAL OF BASIC AND APPLIED SCIENCES ISSN:1991-8178 EISSN: 2309-8414 Journal home page: www.ajbasweb.com Resolution Magnification Technique for Satellite Images Using DT- CWT and NLM 1 Saranya

More information

A DEEP DICTIONARY MODEL FOR IMAGE SUPER-RESOLUTION. Jun-Jie Huang and Pier Luigi Dragotti

A DEEP DICTIONARY MODEL FOR IMAGE SUPER-RESOLUTION. Jun-Jie Huang and Pier Luigi Dragotti A DEEP DICTIONARY MODEL FOR IMAGE SUPER-RESOLUTION Jun-Jie Huang and Pier Luigi Dragotti Communications and Signal Processing Group CSP), Imperial College London, UK ABSTRACT Inspired by the recent success

More information

Bidirectional Recurrent Convolutional Networks for Multi-Frame Super-Resolution

Bidirectional Recurrent Convolutional Networks for Multi-Frame Super-Resolution Bidirectional Recurrent Convolutional Networks for Multi-Frame Super-Resolution Yan Huang 1 Wei Wang 1 Liang Wang 1,2 1 Center for Research on Intelligent Perception and Computing National Laboratory of

More information

IMPROVED MOTION-BASED LOCALIZED SUPER RESOLUTION TECHNIQUE USING DISCRETE WAVELET TRANSFORM FOR LOW RESOLUTION VIDEO ENHANCEMENT

IMPROVED MOTION-BASED LOCALIZED SUPER RESOLUTION TECHNIQUE USING DISCRETE WAVELET TRANSFORM FOR LOW RESOLUTION VIDEO ENHANCEMENT 17th European Signal Processing Conference (EUSIPCO 009) Glasgow, Scotland, August 4-8, 009 IMPROVED MOTION-BASED LOCALIZED SUPER RESOLUTION TECHNIQUE USING DISCRETE WAVELET TRANSFORM FOR LOW RESOLUTION

More information

Edge-Directed Interpolation in a Bayesian Framework

Edge-Directed Interpolation in a Bayesian Framework SIMONYAN, VATOLIN: EDI IN A BAYESIAN FRAMEWORK 1 Edge-Directed Interpolation in a Bayesian Framework Karen Simonyan simonyan@graphics.cs.msu.ru Dmitriy Vatolin dmitriy@yuvsoft.com Graphics & Media Lab

More information

Surface Defect Edge Detection Based on Contourlet Transformation

Surface Defect Edge Detection Based on Contourlet Transformation 2016 3 rd International Conference on Engineering Technology and Application (ICETA 2016) ISBN: 978-1-60595-383-0 Surface Defect Edge Detection Based on Contourlet Transformation Changle Li, Gangfeng Liu*,

More information

Image Super-Resolution Using Local Learnable Kernel Regression

Image Super-Resolution Using Local Learnable Kernel Regression Image Super-Resolution Using Local Learnable Kernel Regression Renjie Liao and Zengchang Qin Intelligent Computing and Machine Learning Lab School of Automation Science and Electrical Engineering Beihang

More information

Learning Parametric Distributions for Image Super-Resolution: Where Patch Matching Meets Sparse Coding

Learning Parametric Distributions for Image Super-Resolution: Where Patch Matching Meets Sparse Coding Learning Parametric Distributions for Image Super-Resolution: Where Patch Matching Meets Sparse Coding Yongbo Li, Weisheng Dong, Guangming Shi, Xuemei Xie School of Electronic Engineering, Xidian University,

More information

Edge Detection. Computer Vision Shiv Ram Dubey, IIIT Sri City

Edge Detection. Computer Vision Shiv Ram Dubey, IIIT Sri City Edge Detection Computer Vision Shiv Ram Dubey, IIIT Sri City Previous two classes: Image Filtering Spatial domain Smoothing, sharpening, measuring texture * = FFT FFT Inverse FFT = Frequency domain Denoising,

More information

Reconstruction PSNR [db]

Reconstruction PSNR [db] Proc. Vision, Modeling, and Visualization VMV-2000 Saarbrücken, Germany, pp. 199-203, November 2000 Progressive Compression and Rendering of Light Fields Marcus Magnor, Andreas Endmann Telecommunications

More information

Efficient Graphical Models for Processing Images

Efficient Graphical Models for Processing Images Abstract Efficient Graphical Models for Processing Images Marshall F. Tappen Bryan C. Russell William T. Freeman Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology

More information

FRESH - An algorithm for resolution enhancement of piecewise smooth signals and images

FRESH - An algorithm for resolution enhancement of piecewise smooth signals and images FRESH - An algorithm for resolution enhancement of piecewise smooth signals and images Imperial College London April 11, 2017 1 1 This research is supported by European Research Council ERC, project 277800

More information

Aggregating Descriptors with Local Gaussian Metrics

Aggregating Descriptors with Local Gaussian Metrics Aggregating Descriptors with Local Gaussian Metrics Hideki Nakayama Grad. School of Information Science and Technology The University of Tokyo Tokyo, JAPAN nakayama@ci.i.u-tokyo.ac.jp Abstract Recently,

More information

EFFICIENT PERCEPTUAL, SELECTIVE,

EFFICIENT PERCEPTUAL, SELECTIVE, EFFICIENT PERCEPTUAL, SELECTIVE, AND ATTENTIVE SUPER-RESOLUTION RESOLUTION Image, Video & Usability (IVU) Lab School of Electrical, Computer, & Energy Engineering Arizona State University karam@asu.edu

More information

Introduction to Image Super-resolution. Presenter: Kevin Su

Introduction to Image Super-resolution. Presenter: Kevin Su Introduction to Image Super-resolution Presenter: Kevin Su References 1. S.C. Park, M.K. Park, and M.G. KANG, Super-Resolution Image Reconstruction: A Technical Overview, IEEE Signal Processing Magazine,

More information

Non-Local Kernel Regression for Image and Video Restoration

Non-Local Kernel Regression for Image and Video Restoration Non-Local Kernel Regression for Image and Video Restoration Haichao Zhang 1,2, Jianchao Yang 2, Yanning Zhang 1, and Thomas S. Huang 2 1 School of Computer Science, Northwestern Polytechnical University,

More information

Face Recognition using SURF Features and SVM Classifier

Face Recognition using SURF Features and SVM Classifier International Journal of Electronics Engineering Research. ISSN 0975-6450 Volume 8, Number 1 (016) pp. 1-8 Research India Publications http://www.ripublication.com Face Recognition using SURF Features

More information