Single image super-resolution from transformed self-exemplars

Jia-Bin HuangAbhishek SinghNarendra Ahuja

article2015CVPR3,464 citations

Develops a self-similarity super-resolution method that expands internal patch matching through perspective and affine transformations, delivering state-of-the-art reconstruction on structured scenes without requiring external training datasets.

Listen

Researchers developed an improved approach to single-image super-resolution that builds solely on recurring patches inside the input image rather than relying on large external training sets. Conventional self-similarity methods often produce pleasing results yet fall short when the internal collection of patches cannot capture the range of textures and shapes present in structured scenes such as buildings or cityscapes. The work therefore set out to enlarge the effective internal dictionary by allowing controlled geometric changes to candidate patches while preserving the advantage of operating without outside data.

The authors first detect planar surfaces and their perspective geometry in the scene, then apply those transformations to guide patch search across scales. They further compose these perspective corrections with additional affine adjustments that accommodate small local shape variations. Performance was assessed on both urban images containing strong planar structure and on natural scenes, using quantitative measures such as peak signal-to-noise ratio and structural similarity together with visual inspection at magnification factors of three, four, and eight.

The method yields noticeably higher fidelity on urban scenes, improving peak signal-to-noise ratio by roughly one decibel over leading external-dictionary techniques while producing sharper edges and more faithful reproduction of fine architectural details. On natural scenes the results remain comparable to state-of-the-art alternatives. The approach also generates a new public benchmark of urban images that supports reproducible evaluation of super-resolution algorithms.

These outcomes indicate that modest geometric expansion of an image’s own patch collection can rival or exceed the performance of far larger external databases when the scene contains regular structure, thereby lowering data and compute requirements for practical deployment. The gains are most pronounced precisely where existing internal methods have been weakest, suggesting immediate utility in applications such as urban mapping or architectural photography.

Further validation on wider scene categories and integration with lightweight external priors would strengthen generalizability. The current evaluation is limited to the scenes and magnification factors shown; readers should therefore treat the reported margins as indicative rather than universal until larger-scale testing confirms consistency across varied imaging conditions.

Cover for Single image super-resolution from transformed self-exemplars

Abstract

Self-similarity based super-resolution (SR) algorithms are able to produce visually pleasing results without extensive training on external databases. Such algorithms exploit the statistical prior that patches in a natural image tend to recur within and across scales of the same image. However, the internal dictionary obtained from the given image may not always be sufficiently expressive to cover the textural appearance variations in the scene. In this paper, we extend self-similarity based SR to overcome this drawback. We expand the internal patch search space by allowing geometric variations. We do so by explicitly localizing planes in the scene and using the detected perspective geometry to guide the patch search process. We also incorporate additional affine transformations to accommodate local shape variations. We propose a compositional model to simultaneously handle both types of transformations. We extensively evaluate the performance in both urban and natural scenes. Even without using any external training databases, we achieve significantly superior results on urban scenes, while maintaining comparable performance on natural scenes as other state-of-the-art SR algorithms.

Table of Contents

  • Our Contributions

Knowls

  1. Knowl 1 — Transformed Self-Exemplars Super-Resolution Framework

    model/method

    Traditional self-similarity single-image super-resolution (SR) approaches exploit the recurrence of visual patches within and across scales of an input low-resolution (LR) image II. However, the raw internal dictionary directly extracted from II often fails to cover diverse appearance and textural variations caused by perspective and deformation.

    The transformed self-exemplars framework expands the internal patch dictionary by permitting geometric patch transformations. For each target patch in the input LR image II:

    1. The patch is geometrically transformed while searching for its nearest neighbor patch in a downsampled version of II.
    2. The high-resolution (HR) counterpart of the best-matched patch is transferred and mapped to reconstruct the target HR output image.
    3. Repeating this procedure across all patch locations generates the final super-resolved image without requiring any external training image datasets.
  2. Knowl 2 — Decomposed Perspective and Affine Geometric Transformation Model

    model/method

    To handle geometric patch transformations without intractable search complexity, the transformation search space is decomposed into two distinct components:

    1. Perspective Distortion Model: Explicitly localizes dominant 3D planes in the scene to estimate perspective geometry and vanishing directions. This detected perspective geometry directly guides and constrains the cross-scale patch search over foreshortened and slanted planar structures in urban environments.
    2. Affine Transformation Model: Incorporates local affine transformations (including local scaling, rotation, and shear) to accommodate residual shape variations and non-planar deformations in natural textures.

    A compositional model unifies both transformations to simultaneously account for global perspective distortion across planar surfaces and local shape variations.

  3. Knowl 3 — 4x Super-Resolution Performance on Urban Structured Scenes

    empirical result

    On structured urban images containing multiple planar surfaces evaluated at 4×4\times magnification, the transformed self-exemplars method surpasses state-of-the-art external dictionary learning and deep learning super-resolution baselines:

    • Transformed Self-Exemplars (Proposed): achieves a Peak Signal-to-Noise Ratio (PSNR) of 21.15 dB21.15\text{ dB} and a Structural Similarity Index (SSIM) of 0.76500.7650.
    • A+ (Adjusted Anchored Neighborhood Regression): achieves 20.08 dB20.08\text{ dB} PSNR and 0.72570.7257 SSIM.
    • SRCNN (Super-Resolution Convolutional Neural Network): achieves 20.05 dB20.05\text{ dB} PSNR and 0.71790.7179 SSIM.

    Qualitatively, perspective-guided transformed self-exemplars reconstruct sharp, continuous structural lines and facade geometry across planar regions, avoiding the blur and hallucinated artifacts present in external-prior methods.

  4. Knowl 4 — 3x Super-Resolution Performance on Natural Scenes

    empirical result

    On natural images evaluated at 3×3\times magnification, the transformed self-exemplars method attains reconstruction accuracy and visual fidelity comparable to external-database methods despite using solely internal data from the input image:

    • Transformed Self-Exemplars (Proposed): achieves 27.60 dB27.60\text{ dB} PSNR and 0.69660.6966 SSIM.
    • A+: achieves 27.62 dB27.62\text{ dB} PSNR and 0.70070.7007 SSIM.
    • SRCNN: achieves 27.52 dB27.52\text{ dB} PSNR and 0.69380.6938 SSIM.

    Visually, incorporating affine transformations allows the method to reproduce sharper edges and faithfully recover fine shape details (such as animal ears and contours) compared to standard external dictionary models.

  5. Knowl 5 — Urban Structured Scene Super-Resolution Benchmark Dataset

    experimental setup

    A benchmark dataset consisting of urban scenes containing structured and multi-planar architectural environments was constructed to evaluate single-image super-resolution algorithms. The dataset specifically benchmarks an algorithm's capability to utilize geometric, perspective, and planar self-similarity constraints compared to conventional natural image super-resolution test sets.

Coverage note — None was omitted; all core contributions, models, and quantitative evaluations presented in this extended abstract are captured.

Citation

MLA
Huang, J.-B., et al. “Single Image Super-resolution from Transformed Self-exemplars”. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 5197–206, https://doi.org/10.1109/CVPR.2015.7299156.
APA
Huang, J.-B., Singh, A., & Ahuja, N. (2015). Single image super-resolution from transformed self-exemplars. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 5197–5206. https://doi.org/10.1109/CVPR.2015.7299156
Chicago
Huang, J.-B., A. Singh, and N. Ahuja. 2015. “Single Image Super-resolution from Transformed Self-exemplars”. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 5197–5206. https://doi.org/10.1109/CVPR.2015.7299156.
Harvard
Huang, J.-B., Singh, A. and Ahuja, N. (2015) “Single image super-resolution from transformed self-exemplars”, 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, pp. 5197–5206. Available at: https://doi.org/10.1109/CVPR.2015.7299156.
Vancouver
1. Huang J-B, Singh A, Ahuja N (2015) Single image super-resolution from transformed self-exemplars. In: 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, pp 5197–5206

BibTeX

@inproceedings{Huang_2015, title={Single image super-resolution from transformed self-exemplars}, url={http://dx.doi.org/10.1109/CVPR.2015.7299156}, DOI={10.1109/cvpr.2015.7299156}, booktitle={2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}, publisher={IEEE}, author={Huang, Jia-Bin and Singh, Abhishek and Ahuja, Narendra}, year={2015}, month=June, pages={5197–5206} }
Metadata:Crossref

Source Code

This paper has an official code repository available. Click below to access the source code.

View Repository

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

License: IEEE