Single-frame Super-resolution Research Articles

Multi-frame image super-resolution reconstruction using sparse co-occurrence prior and sub-pixel registration

In this paper, a hybrid super-resolution (SR) method is proposed by combining the concepts of both multi-frame and single-frame SR to generate a high-resolution (HR) image. The main contributions are in two aspects: the first one is hierarchical iterative sub-pixel registration, which provides accurate registration information of input low-resolution (LR) images or frames, to generate an initial HR image; the second one is to enhance the initial HR image with sparse co-occurrence prior, resulted from specially-designed dictionaries containing patches from both generic training images and interpolated input LR images. As a whole, the proposed hybrid SR method makes use of information from both sub-pixel registration and sparse co-occurrence prior to get reconstructed SR image with large zoom-in factor. The simulation results from synthetic images and real video frames illustrate its effectiveness and the superiority in image quality over conventional multi-frame and single-frame SR methods.

Fast Example-Based Super-Resolution Using Manifold Learning

This paper presents a new method for single-frame Super-Resolution (SR), by combining Example-based SR and neighbor embedding based SR (NE-based SR). Example-based SR attempts to generate High-Resolution (HR) image through estimating the High-Frequency (HF) components that are lost in the input Low-Resolution (LR) image. This method usually can achieve acceptable HR images if enough amounts of similar training samples are prepared. However, the HF component is approximated by only one training sample, which easily produces noise and artifacts. On the other hand, NE-based SR recovers HR image using manifold learning - Locally Linear Embedding, which represents any LR input and its corresponding HR one by a weighted linear combination of several training patches. The NE-based SR need to prepare large-scale training database with both intensity and structure variation, which will lead to high computation. This study combines these two methods to only estimate the HF components using several training samples. Moreover, we extend the proposed method to a fast version by processing only the patches with large variance. Experimental results show that the reconstructed HR images by our proposed approach are much better than those by conventional methods and interpolation techniques, and at the same time the computation is much faster.

Image superresolution using fractal coding

We address the problem of image superresolution and present a novel approach to single-frame superresolution using fractal image coding. The proposed approach takes great advantage of the properties of fractals—resolution independence, similarity preservation, and nonlinear operation—which are suited for image superresolution, specifically for image restoration and magnification. The idea of our work is to estimate the fractal code of the original image from its degraded blurred and noised observation and decode it at a higher resolution, and all the strategies are performed in the fractal image coding frame- work. To achieve this, we employ an adaptive fractal coding scheme in the frequency domain, and further, we introduce an overlapping partition scheme to remove the blocky artifacts and improve the reconstruction quality. Experiments on simulated and real images show that the result- ing fractal-based superresolution method yields superior performance to conventional single-frame superresolution methods. © 2008 Society of Photo-

Joint Blind Super-Resolution and Shadow Removing

Most learning-based super-resolution methods neglect the illumination problem. In this paper we propose a novel method to combine blind single-frame super-resolution and shadow removal into a single operation. Firstly, from the pattern recognition viewpoint, blur identification is considered as a classification problem. We describe three methods which are respectively based on Vector Quantization (VQ), Hidden Markov Model (HMM) and Support Vector Machines (SVM) to identify the blur parameter of the acquisition system from the compressed/uncompressed low-resolution image. Secondly, after blur identification, a super-resolution image is reconstructed by a learning-based method. In this method, Logarithmic-wavelet transform is defined for illumination-free feature extraction. Then an initial estimation is obtained based on the assumption that small patches in low-resolution space and patches in high-resolution space share a similar local manifold structure. The unknown high-resolution image is reconstructed by projecting the intermediate result into general reconstruction constraints. The proposed method simultaneously achieves blind single-frame super-resolution and image enhancement especially shadow removal. Experimental results demonstrate the effectiveness and robustness of our method.

Iterative kernel principal component analysis for image modeling

In recent years, Kernel Principal Component Analysis (KPCA) has been suggested for various image processing tasks requiring an image model such as, e.g., denoising or compression. The original form of KPCA, however, can be only applied to strongly restricted image classes due to the limited number of training examples that can be processed. We therefore propose a new iterative method for performing KPCA, the Kernel Hebbian Algorithm which iteratively estimates the Kernel Principal Components with only linear order memory complexity. In our experiments, we compute models for complex image classes such as faces and natural images which require a large number of training examples. The resulting image models are tested in single-frame super-resolution and denoising applications. The KPCA model is not specifically tailored to these tasks; in fact, the same model can be used in super-resolution with variable input resolution, or denoising with unknown noise characteristics. In spite of this, both super-resolution and denoising performance are comparable to existing methods.