Overcoming Registration Uncertainty in Image Super-Resolution: Maximize or Marginalize?
- PDF / 2,359,765 Bytes
- 14 Pages / 612 x 792 pts (letter) Page_size
- 41 Downloads / 157 Views
Research Article Overcoming Registration Uncertainty in Image Super-Resolution: Maximize or Marginalize? Lyndsey C. Pickup, David P. Capel, Stephen J. Roberts, and Andrew Zisserman Information Engineering Building, Department of Engineering Science, Parks Road, Oxford OX1 3PJ, UK Received 15 September 2006; Accepted 4 May 2007 Recommended by Russell C. Hardie In multiple-image super-resolution, a high-resolution image is estimated from a number of lower-resolution images. This usually involves computing the parameters of a generative imaging model (such as geometric and photometric registration, and blur) and obtaining a MAP estimate by minimizing a cost function including an appropriate prior. Two alternative approaches are examined. First, both registrations and the super-resolution image are found simultaneously using a joint MAP optimization. Second, we perform Bayesian integration over the unknown image registration parameters, deriving a cost function whose only variables of interest are the pixel values of the super-resolution image. We also introduce a scheme to learn the parameters of the image prior as part of the super-resolution algorithm. We show examples on a number of real sequences including multiple stills, digital video, and DVDs of movies. Copyright © 2007 Lyndsey C. Pickup et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
1.
INTRODUCTION
Multiframe image super-resolution refers to the process by which a set of images of the same scene are fused to produce an image or images with a higher spatial resolution, or with more visible detail in the high spatial frequency features [1]. The limits on the resolution of the original imaging device can be improved by exploiting the relative subpixel motion between the scene and the imaging plane. Applications are common, with everything from holiday snaps and DVD frames to satellite terrain imagery providing collections of low-resolution images to be enhanced, for instance to produce a more aesthetic image for media publication [2, 3], object or surface reconstruction [4], or for higher-level vision tasks such as object recognition or localization [5]. Figure 1 shows examples from a still camera and a DVD movie. In previous work, a few methods have assumed no scene motion, and use other cues such as lighting or varying zoom [6]. However, the vast majority of current super-resolution methods do assume motion, and either preregister the inputs using standard registration techniques, or assume that a perfect registration is given a priori [1, 7], before carrying out the super-resolution estimate. However, the steps taken in super-resolution are seldom truly independent, and this is too often ignored in current super-resolution techniques
[1, 7–12]. In this work we will develop two algorithms which consider the problem in a more unified way. The first approach is to estimate a super-resol
Data Loading...
