Determining the regularization parameters for super-resolution problems

Zibetti, Marcelo V. W.; Bazán, Fermín S. V.; Mayer, Joceli

doi:10.1016/j.sigpro.2008.06.010

Cited by 40 publications

(27 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Various optimization approaches were proposed to further stabilize the inversion of such an ill-posed problem, such as [8], [9], [11]. While, the performance of these algorithms degrades rapidly when the desired magnification factor is large or the number of input images is small.…”

Section: Existing Workmentioning

confidence: 99%

“…Over the last few years, numerous super resolution methods have been developed and applied to a variety of image classes with differing degrees of success on different image classes (e.g. [8], [9], [10], [11], [12], [13], [14], [15], [16]). In this paper, we investigate two existing spatial domain super resolution methods to reconstruct high-resolution face images from single/multiple low-resolution images primarily to test their suitability for face recognition at a distance.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Enhancing face recognition at a distance using super resolution

Al-Hassan

Jassim

Sellahewa

2012

Proceedings of the on Multimedia and Security

View full text Add to dashboard Cite

The characteristics of surveillance video generally include lowresolution images and blurred images. Decreases in image resolution leads to loss of high frequency facial components, which is expected to adversely affect recognition rates. Super resolution (SR) is a technique used to generate a higher resolution image from a given low-resolution, degraded image. In this paper, we investigate two existing super-resolution methods to reconstruct a high resolution from a single/ multiple lowresolution images. The first method is based on pairs of high and low quality dictionaries in the spatial domain and the other method is the so called Back-Project Iterative Interpolation for super resolution. Recent works have used SR as a pre-processing step to overcome the problem of low-resolution images in face recognition. However, super resolution reconstruction process being ill-posed results in visual artifacts that can be visually distracting to humans and/or effect machine feature extraction and face recognition algorithms. In this paper, we propose an alternative super resolution scheme based on dictionaries in high frequency wavelet subbands. The proposed method is evaluated on database of high and low-resolution images-Extended Yale B database and UBHSD database-and for face recognition at a distance. We shall demonstrate that the proposed approach at level 3 DWT resolution has superior performance in comparison to the other methods.

show abstract

Section: Existing Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Enhancing face recognition at a distance using super resolution

Al-Hassan

Jassim

Sellahewa

2012

Proceedings of the on Multimedia and Security

View full text Add to dashboard Cite

show abstract

“…While automating this choice has been successfully tackled for G-MRF [2], the same is currently not available for H-MRF. Due to the intractable form of the partition function of Huber-MRF, a principled model-based estimation of the hyper-parameters has been found problematic [7], and various heuristic approaches have been in use [4,7] to sidestep the computational burden of crossvalidation.…”

Section: Introductionmentioning

confidence: 99%

Estimation of the Regularisation Parameter in Huber-MRF for Image Resolution Enhancement

Pitchay

Kabán

2013

Intelligent Data Engineering and Automated Learning – IDEAL 2013

View full text Add to dashboard Cite

Abstract. The Huber Markov Random Field (H-MRF) has been proposed for image resolution enhancement as a preferable alternative to Gaussian Random Markov Fields (G-MRF) for its ability to preserve discontinuities in the image. However, its performance relies on a good choice of a regularisation parameter. While automating this choice has been successfully tackled for G-MRF, the more sophisticated form of H-MRF makes this problem less straightforward. In this paper we develop an approximate solution to this problem, by upper-bounding the partition function of the H-MRF. We demonstrate the working and flexibility of our approach in image super-resolution experiments.

show abstract

“…SR image reconstruction is generally a severely ill-posed problem because of the insufficient number of LR images, ill-conditioned registration, unknown blurring operator and the solution from the reconstruction constraint is not unique. Therefore, various regularization techniques are proposed to stabilize the inversion of this ill-posed problem [1], [2], [3]. However, the performance of these algorithms degrades rapidly when the desired magnification factor is large or the number of the input images is small.…”

Section: Introductionmentioning

confidence: 99%