A Bayesian approach to image expansion for improved definition

Projection-based image registration in the presence of fixed-pattern noise

A computationally efficient method for image registration is investigated that can achieve an improved performance over the traditional two-dimensional (2-D) cross-correlation-based techniques in the presence of both fixed-pattern and temporal noise. The method relies on transforming each image in the sequence of frames into two vector projections formed by accumulating pixel values along the rows and columns of the image. The vector projections corresponding to successive frames are in turn used to estimate the individual horizontal and vertical components of the shift by means of a one-dimensional (1-D) cross-correlation-based estimator. While gradient-based shift estimation techniques are computationally efficient, they often exhibit degraded performance under noisy conditions in comparison to cross-correlators due to the fact that the gradient operation amplifies noise. The projection-based estimator, on the other hand, significantly reduces the computational complexity associated with the 2-D operations involved in traditional correlation-based shift estimators while improving the performance in the presence of temporal and spatial noise. To show the noise rejection capability of the projection-based shift estimator relative to the 2-D cross correlator, a figure-of-merit is developed and computed reflecting the signal-to-noise ratio (SNR) associated with each estimator. The two methods are also compared by means of computer simulation and tests using real image sequences.

A filter design method for minimizing ringing in a region of interest in MR spectroscopic images

Magnetic resonance spectroscopic imaging (MRSI) requires a relatively long time to sample k-space (the spatial frequency domain), effectively lowpass filtering the resulting reconstructed image. Ringing is especially problematic when a region of interest (ROI) is close to a bright region outside the ROI, since the bright region tends to create a ringing artifact into the ROI due to the lowpass nature of the data. In this paper, we propose a method that reduces the effect of a stronger signal region on a weaker signal in a nearby ROI by designing a postprocessing filter that steers the strong interference away from the ROI. The proposed method is computationally simple both in the design stage and in applying it to images. We present experiments that illustrate the value of the technique.

Survey: interpolation methods in medical image processing

Image interpolation techniques often are required in medical imaging for image generation (e.g., discrete back projection for inverse Radon transform) and processing such as compression or resampling. Since the ideal interpolation function spatially is unlimited, several interpolation kernels of finite size have been introduced. This paper compares 1) truncated and windowed sinc; 2) nearest neighbor; 3) linear; 4) quadratic; 5) cubic B-spline; 6) cubic; g) Lagrange; and 7) Gaussian interpolation and approximation techniques with kernel sizes from 1 x 1 up to 8 x 8. The comparison is done by: 1) spatial and Fourier analyses; 2) computational complexity as well as runtime evaluations; and 3) qualitative and quantitative interpolation error determinations for particular interpolation tasks which were taken from common situations in medical image processing. For local and Fourier analyses, a standardized notation is introduced and fundamental properties of interpolators are derived. Successful methods should be direct current (DC)-constant and interpolators rather than DC-inconstant or approximators. Each method's parameters are tuned with respect to those properties. This results in three novel kernels, which are introduced in this paper and proven to be within the best choices for medical image interpolation: the 6 x 6 Blackman-Harris windowed sinc interpolator, and the C2-continuous cubic kernels with N = 6 and N = 8 supporting points. For quantitative error evaluations, a set of 50 direct digital X rays was used. They have been selected arbitrarily from clinical routine. In general, large kernel sizes were found to be superior to small interpolation masks. Except for truncated sinc interpolators, all kernels with N = 6 or larger sizes perform significantly better than N = 2 or N = 3 point methods (p << 0.005). However, the differences within the group of large-sized kernels were not significant. Summarizing the results, the cubic 6 x 6 interpolator with continuous second derivatives, as defined in (24), can be recommended for most common interpolation tasks. It appears to be the fastest six-point kernel to implement computationally. It provides eminent local and Fourier properties, is easy to implement, and has only small errors. The same characteristics apply to B-spline interpolation, but the 6 x 6 cubic avoids the intrinsic border effects produced by the B-spline technique. However, the goal of this study was not to determine an overall best method, but to present a comprehensive catalogue of methods in a uniform terminology, to define general properties and requirements of local techniques, and to enable the reader to select that method which is optimal for his specific application in medical imaging.

Regularity-preserving image interpolation

Assumptions about image continuity lead to oversmoothed edges in common image interpolation algorithms. A wavelet-based interpolation method that imposes no continuity constraints is introduced. The algorithm estimates the regularity of edges by measuring the decay of wavelet transform coefficients across scales and preserves the underlying regularity by extrapolating a new subband to be used in image resynthesis. The algorithm produces visibly sharper edges than traditional techniques and exhibits an average peak signal-to-noise ratio (PSNR) improvement of 2.5 dB over bilinear and bicubic techniques.

Warped distance for space-variant linear image interpolation

The problem of image interpolation using linear techniques is dealt with in this paper. Conventional space-invariant methods are revisited and changed into space-variant ones, by introducing the concept of the warped distance among the pixels of an image. A better perceptual rendition of the image details is obtained in this way; this effect is proved both via the evaluation of the response to an idealized sigmoidal edge model and with experiments on real-world images. The computational costs of the proposed approach are very small when compared to those of state-of-the art nonlinear interpolation operators.

Fast registration and reconstruction of aliased low-resolution frames by use of a modified maximum-likelihood approach

During the process of microscanning a controlled vibrating mirror typically is used to produce subpixel shifts in a sequence of forward-looking infrared (FLIR) images. If the FLIR is mounted on a moving platform, such as an aircraft, uncontrolled random vibrations associated with the platform can be used to generate the shifts. Iterative techniques such as the expectation-maximization (EM) approach by means of the maximum-likelihood algorithm can be used to generate high-resolution images from multiple randomly shifted aliased frames. In the maximum-likelihood approach the data are considered to be Poisson random variables and an EM algorithm is developed that iteratively estimates an unaliased image that is compensated for known imager-system blur while it simultaneously estimates the translational shifts. Although this algorithm yields high-resolution images from a sequence of randomly shifted frames, it requires significant computation time and cannot be implemented for real-time applications that use the currently available high-performance processors. The new image shifts are iteratively calculated by evaluation of a cost function that compares the shifted and interlaced data frames with the corresponding values in the algorithm's latest estimate of the high-resolution image. We present a registration algorithm that estimates the shifts in one step. The shift parameters provided by the new algorithm are accurate enough to eliminate the need for iterative recalculation of translational shifts. Using this shift information, we apply a simplified version of the EM algorithm to estimate a high-resolution image from a given sequence of video frames. The proposed modified EM algorithm has been found to reduce significantly the computational burden when compared with the original EM algorithm, thus making it more attractive for practical implementation. Both simulation and experimental results are presented to verify the effectiveness of the proposed technique.

Globally convergent edge-preserving regularized reconstruction: an application to limited-angle tomography

We introduce a generalization of a deterministic relaxation algorithm for edge-preserving regularization in linear inverse problems. This algorithm transforms the original (possibly nonconvex) optimization problem into a sequence of quadratic optimization problems, and has been shown to converge under certain conditions when the original cost functional being minimized is strictly convex. We prove that our more general algorithm is globally convergent (i.e., converges to a local minimum from any initialization) under less restrictive conditions, even when the original cost functional is nonconvex. We apply this algorithm to tomographic reconstruction from limited-angle data by formulating the problem as one of regularized least-squares optimization. The results demonstrate that the constraint of piecewise smoothness, applied through the use of edge-preserving regularization, can provide excellent limited-angle tomographic reconstructions. Two edge-preserving regularizers-one convex, the other nonconvex-are used in numerous simulations to demonstrate the effectiveness of the algorithm under various limited-angle scenarios, and to explore how factors, such as the choice of error norm, angular sampling rate and amount of noise, affect the reconstruction quality and algorithm performance. These simulation results show that for this application, the nonconvex regularizer produces consistently superior results.

Joint MAP registration and high-resolution image estimation using a sequence of undersampled images

In many imaging systems, the detector array is not sufficiently dense to adequately sample the scene with the desired field of view. This is particularly true for many infrared focal plane arrays. Thus, the resulting images may be severely aliased. This paper examines a technique for estimating a high-resolution image, with reduced aliasing, from a sequence of undersampled frames. Several approaches to this problem have been investigated previously. However, in this paper a maximum a posteriori (MAP) framework for jointly estimating image registration parameters and the high-resolution image is presented. Several previous approaches have relied on knowing the registration parameters a priori or have utilized registration techniques not specifically designed to treat severely aliased images. In the proposed method, the registration parameters are iteratively updated along with the high-resolution image in a cyclic coordinate-descent optimization procedure. Experimental results are provided to illustrate the performance of the proposed MAP algorithm using both visible and infrared images. Quantitative error analysis is provided and several images are shown for subjective evaluation.

Superresolution video reconstruction with arbitrary sampling lattices and nonzero aperture time

Printing from an NTSC source and conversion of NTSC source material to high-definition television (HDTV) format are some of the applications that motivate superresolution (SR) image and video reconstruction from low-resolution (LR) and possibly blurred sources. Existing methods for SR image reconstruction are limited by the assumptions that the input LR images are sampled progressively, and that the aperture time of the camera is zero, thus ignoring the motion blur occurring during the aperture time. Because of the observed adverse effects of these assumptions for many common video sources, this paper proposes (i) a complete model of video acquisition with an arbitrary input sampling lattice and a nonzero aperture time, and (ii) an algorithm based on this model using the theory of projections onto convex sets to reconstruct SR still images or video from an LR time sequence of images. Experimental results with real video are provided, which clearly demonstrate that a significant increase in the image resolution can be achieved by taking the motion blurring into account especially when there exists large interframe motion.

A fast and accurate Fourier algorithm for iterative parallel-beam tomography

We use a series-expansion approach and an operator framework to derive a new, fast, and accurate Fourier algorithm for iterative tomographic reconstruction. This algorithm is applicable for parallel-ray projections collected at a finite number of arbitrary view angles and radially sampled at a rate high enough that aliasing errors are small. The conjugate gradient (CG) algorithm is used to minimize a regularized, spectrally weighted least-squares criterion, and we prove that the main step in each iteration is equivalent to a 2-D discrete convolution, which can be cheaply and exactly implemented via the fast Fourier transform (FFT). The proposed algorithm requires O(N(2)logN) floating-point operations per iteration to reconstruct an NxN image from P view angles, as compared to O(N (2)P) floating-point operations per iteration for iterative convolution-backprojection algorithms or general algebraic algorithms that are based on a matrix formulation of the tomography problem. Numerical examples using simulated data demonstrate the effectiveness of the algorithm for sparse- and limited-angle tomography under realistic sampling scenarios. Although the proposed algorithm cannot explicitly account for noise with nonstationary statistics, additional simulations demonstrate that for low to moderate levels of nonstationary noise, the quality of reconstruction is almost unaffected by assuming that the noise is stationary.

Extraction of high-resolution frames from video sequences

The human visual system appears to be capable of temporally integrating information in a video sequence in such a way that the perceived spatial resolution of a sequence appears much higher than the spatial resolution of an individual frame. While the mechanisms in the human visual system that do this are unknown, the effect is not too surprising given that temporally adjacent frames in a video sequence contain slightly different, but unique, information. This paper addresses the use of both the spatial and temporal information present in a short image sequence to create a single high-resolution video frame. A novel observation model based on motion compensated subsampling is proposed for a video sequence. Since the reconstruction problem is ill-posed, Bayesian restoration with a discontinuity-preserving prior image model is used to extract a high-resolution video still given a short low-resolution sequence. Estimates computed from a low-resolution image sequence containing a subpixel camera pan show dramatic visual and quantitative improvements over bilinear, cubic B-spline, and Bayesian single frame interpolations. Visual and quantitative improvements are also shown for an image sequence containing objects moving with independent trajectories. Finally, the video frame extraction algorithm is used for the motion-compensated scan conversion of interlaced video data, with a visual comparison to the resolution enhancement obtained from progressively scanned frames.

Interpolation of missing data in image sequences

This paper presents a number of model based interpolation schemes tailored to the problem of interpolating missing regions in image sequences. These missing regions may be of arbitrary size and of random, but known, location. This problem occurs regularly with archived film material. The film is abraded or obscured in patches, giving rise to bright and dark flashes, known as "dirt and sparkle" in the motion picture industry. Both 3-D autoregressive models and 3-D Markov random fields are considered in the formulation of the different reconstruction processes. The models act along motion directions estimated using a multiresolution block matching scheme. It is possible to address this sort of impulsive noise suppression problem with median filters, and comparisons with earlier work using multilevel median filters are performed. These comparisons demonstrate the higher reconstruction fidelity of the new interpolators.

Stochastic modeling and estimation of multispectral image data

Multispectral images consist of multiple channels, each containing data acquired from a different band within the frequency spectrum. Since most objects emit or reflect energy over a large spectral bandwidth, there usually exists a significant correlation between channels. Due to often harsh imaging environments, the acquired data may be degraded by both blur and noise. Simply applying a monochromatic restoration algorithm to each frequency band ignores the cross-channel correlation present within a multispectral image. A Gibbs prior is proposed for multispectral data modeled as a Markov random field, containing both spatial and spectral cliques. Spatial components of the model use a nonlinear operator to preserve discontinuities within each frequency band, while spectral components incorporate nonstationary cross-channel correlations. The multispectral model is used in a Bayesian algorithm for the restoration of color images, in which the resulting nonlinear estimates are shown to be quantitatively and visually superior to linear estimates generated by multichannel Wiener and least squares restoration.