Diagnosis of hip osteoarthritis is conventionally done through a manual measurement of the joint distance between the femoral head and the acetabular cup, a difficult and often error-prone process. Recently, Chen et al.1 proposed a fully automated technique based on landmark displacement estimation from multiple image patches that is able to accurately segment bone structures around the pelvis. This technique was shown to be comparable or better than state-of-the-art random forest based methods. In this paper, we report on the implementation and evaluation of this method on low-resolution datasets typically available in parts of the developing world where high-resolution X-ray image technology is unavailable.
We employed a dataset of hip joint images collected at a local clinic and provided to us in JPG format and at 1/3 the resolution of typical DICOM X-ray images. In addition, we employed the Dice similarity coefficient, average Euclidean distance between corresponding landmarks, and Hausdorff distance to better evaluate the method relative to diagnosis of hip osteoarthritis. Our results show that the proposed method is robust with JPEG images at 1/3 the resolution of DICOM data. Additional preliminary results quantify the accuracy of the approach as a function of decreasing resolution. We believe these results have important significance for application in clinical settings where modern X-ray equipment is not available.
We evaluate recently developed randomized matrix decomposition methods for fast lossless compression and reconstruction of hyperspectral imaging (HSI) data. The simple random projection methods have been shown to be effective for lossy compression without severely affecting the performance of object identification and classification. We build upon these methods to develop a new double-random projection method that may enable security in data transmission of compressed data. For HSI data, the distribution of elements in the resulting residual matrix, i.e., the original data subtracted by its low-rank representation, exhibits a low entropy relative to the original data that favors high-compression ratio. We show both theoretically and empirically that randomized methods combined with residual-coding algorithms can lead to effective lossless compression of HSI data. We conduct numerical tests on real large-scale HSI data that shows promise in this case. In addition, we show that randomized techniques can be applicable for encoding on resource-constrained on-board sensor systems, where the core matrix-vector multiplications can be easily implemented on computing platforms such as graphic processing units or field-programmable gate arrays.
Ocular recognition is a new area of biometric investigation targeted at overcoming the limitations of iris recognition
performance in the presence of non-ideal data. There are several advantages for increasing the area beyond
the iris, yet there are also key issues that must be addressed such as size of the ocular region, factors affecting
performance, and appropriate corpora to study these factors in isolation. In this paper, we explore and identify
some of these issues with the goal of better defining parameters for ocular recognition. An empirical study is
performed where iris recognition methods are contrasted with texture and point operators on existing iris and
face datasets. The experimental results show a dramatic recognition performance gain when additional features
are considered in the presence of poor quality iris data, offering strong evidence for extending interest beyond
the iris. The experiments also highlight the need for the direct collection of additional ocular imagery.
An integrated array computational imaging system, dubbed PERIODIC, is presented which is capable of exploiting a
diverse variety of optical information including sub-pixel displacements, phase, polarization, intensity, and
wavelength. Several applications of this technology will be presented including digital superresolution, enhanced
dynamic range and multi-spectral imaging. Other applications include polarization based dehazing, extended depth of
field and 3D imaging. The optical hardware system and software algorithms are described, and sample results are
Digital super-resolution refers to computational techniques that exploit the generalized sampling theorem to
extend image resolution beyond the pixel spacing of the detector, but not beyond the optical limit (Nyquist
spatial frequency) of the lens. The approach to digital super-resolution taken by the PERIODIC multi-lenslet
camera project is to solve a forward model which describes the effects of sub-pixel shifts, optical blur, and
detector sampling as a product of matrix factors. The associated system matrix is often ill-conditioned, and
convergence of iterative methods to solve for the high-resolution image may be slow.
We investigate the use of pupil phase encoding in a multi-lenslet camera system as a means to physically
precondition and regularize the computational super-resolution problem. This is an integrated optical-digital
approach that has been previously demonstrated with cubic type and pseudo-random phase elements. Traditional
multi-frame phase diversity for imaging through atmospheric turbulence uses a known smooth phase perturbation
to help recover a time series of point spread functions corresponding to random phase errors. In the context of a
multi-lenslet camera system, a known pseudo-random or cubic phase error may be used to help recover an array
of unknown point spread functions corresponding to manufacturing and focus variations among the lenslets.
We investigate the use of a novel multi-lens imaging system in the context of biometric identification, and more
specifically, for iris recognition. Multi-lenslet cameras offer a number of significant advantages over standard
single-lens camera systems, including thin form-factor and wide angle of view. By using appropriate lenslet spacing
relative to the detector pixel pitch, the resulting ensemble of images implicitly contains subject information
at higher spatial frequencies than those present in a single image. Additionally, a multi-lenslet approach enables
the use of observational diversity, including phase, polarization, neutral density, and wavelength diversities. For
example, post-processing multiple observations taken with differing neutral density filters yields an image having
an extended dynamic range. Our research group has developed several multi-lens camera prototypes for the
investigation of such diversities.
In this paper, we present techniques for computing a high-resolution reconstructed image from an ensemble of
low-resolution images containing sub-pixel level displacements. The quality of a reconstructed image is measured
by computing the Hamming distance between the Daugman4 iris code of a conventional reference iris image,
and the iris code of a corresponding reconstructed image. We present numerical results concerning the effect of
noise and defocus blur in the reconstruction process using simulated data and report preliminary work on the
reconstruction of actual iris data obtained with our camera prototypes.
Iris recognition imaging is attracting considerable interest as a viable alternative for personal identification and verification in many defense and security applications. However current iris recognition systems suffer from limited depth of field, which makes usage of these systems more difficult by an untrained user. Traditionally, the depth of field is increased by reducing the imaging system aperture, which adversely impacts the light capturing power and thus the system signal-to-noise ratio (SNR). In this paper we discuss a computational imaging system, referred to as Wavefront Coded(R) imaging, for increasing the depth of field without sacrificing the SNR or the resolution of the imaging system. This system employs a especially designed Wavefront Coded lens customized for iris recognition. We present experimental results that show the benefits of this technology for biometric identification.
The insertion of a suitably designed phase plate in the pupil of an imaging system makes it possible to encode the depth dimension of an extended three-dimensional scene by means of an approximately shift-invariant PSF. The so-encoded image can then be deblurred digitally by standard image recovery algorithms to recoup the depth dependent detail of the original scene. A similar strategy can be adopted to compensate for certain monochromatic aberrations of the system. Here we consider two approaches to optimizing the design of the phase plate that are somewhat complementary - one based on Fisher information that attempts to reduce the sensitivity of the phase encoded image to misfocus and the other based on a minimax formulation of the sum of singular values of the system blurring matrix that attempts to maximize the resolution in the final image. Comparisons of these two optimization approaches are discussed. Our preliminary demonstration of the use of such pupil-phase engineering to successfully control system aberrations, particularly spherical aberration, is also presented.
Computational imaging systems are modern systems that consist of generalized aspheric optics and image processing capability. These systems can be optimized to greatly increase the performance above systems consisting solely of traditional optics. Computational imaging technology can be used to advantage in iris recognition applications. A major difficulty in current iris recognition systems is a very shallow depth-of-field that limits system usability and increases system complexity. We first review some current iris recognition algorithms, and then describe computational imaging approaches to iris recognition using cubic phase wavefront encoding. These new approaches can greatly increase the depth-of-field over that possible with traditional optics, while keeping sufficient recognition accuracy. In these approaches the combination of optics, detectors, and image processing all contribute to the iris recognition accuracy and efficiency. We describe different optimization methods for designing the optics and the image processing algorithms, and provide laboratory and simulation results from applying these systems and results on restoring the intermediate phase encoded images using both direct Wiener filter and iterative conjugate gradient methods.
Automated iris recognition is a promising method for noninvasive verification of identity. Although it is noninvasive, the procedure requires considerable cooperation from the user. In typical acquisition systems, the subject must carefully position the head laterally to make sure that the captured iris falls within the field-of-view of the digital image acquisition system. Furthermore, the need for sufficient energy at the plane of the detector calls for a relatively fast optical system which results in a narrow depth-of-field. This latter issue requires the user to move the head back and forth until the iris is in good focus. In this paper, we address the depth-of-field problem by studying the effectiveness of specially designed aspheres that extend the depth-of-field of the image capture system. In this initial study, we concentrate on the cubic phase mask originally proposed by Dowski and Cathey. Laboratory experiments are used to produce representative captured irises with and without cubic asphere masks modifying the imaging system. The iris images are then presented to a well-known iris recognition algorithm proposed by Daugman. In some cases we present unrestored imagery and in other cases we attempt to restore the moderate blur introduced by the asphere. Our initial results show that the use of such aspheres does indeed relax the depth-of-field requirements even without restoration of the blurred images. Furthermore, we find that restorations that produce visually pleasing iris images often actually degrade the performance of the algorithm. Different restoration parameters are examined to determine their usefulness in relation to the recognition algorithm.
A novel and successful optical-digital approach for removing certain
aberrations in imaging systems involves placing an optical mask between an image-recording device and an object to encode the wavefront phase before the image is recorded, followed by digital image deconvolution to decode the phase. We have observed that when appropriately engineered, such an optical mask can also act as a form of preconditioner for certain deconvolution algorithms. It can boost information in the signal before it is recorded well above the noise level, leveraging digital restorations of very high quality. In this paper, we 1) examine the influence that a phase mask has on the incoming signal and how it subsequently affects the performance of restoration algorithms, and 2) explore the design of optical masks, a difficult nonlinear optimization problem with multiple design parameters, for removing certain aberrations and for maximizing
restorability and information in recorded images.
By suitably phase-encoding optical images in the pupil plane and then digitally restoring them, one can greatly improve their quality. The use of a cubic phase mask originated by Dowski and Cathey to enhance the depth of focus in the images of 3-d scenes is a classic example of this powerful approach. By using the Strehl ratio as a measure of image quality, we propose tailoring the pupil phase profile by minimizing the sensitivity of the quality of the phase-encoded image of a point source to both its lateral and longitudinal coordinates. Our approach ensures that the encoded image will be formed under a nearly shift-invariant imaging condition, which can then be digitally restored to a high overall quality nearly free from the aberrations and limited depth of focus of a traditional imaging system. We also introduce an alternative measure of sensitivity that is based on the concept of Fisher information. In order to demonstrate the validity of our general approach, we present results of computer simulations that include the limitations imposed by detector noise.
Many visible and infrared sampled imaging systems suffer from moderate to severe amounts of aliasing. The problem arises because the large optical apertures required for sufficient light gathering ability result in large spatial cutoff frequencies. In consumer grade cameras, images are often undersampled by a factor of twenty times the suggested Nyquist rate. Most consumer cameras employ birefringent blur filters that purposely blur the image prior to detection to reduce Moire artifacts produced by aliasing. In addition to the obvious Moire artifacts, aliasing introduces other pixel level errors that can cause artificial jagged edges and erroneous intensity values. These types of errors have led some investigators to treat the aliased signal as noise in imaging system design and analysis. The importance of aliasing is dependent on the nature of the imagery and the definition of the assessment task. In this study, we employ a laboratory experiment to characterize the nature of aliasing noise for a variety of object classes. We acquire both raw and blurred imagery to explore the impact of pre-detection antialiasing. We also consider the post detection image restoration requirements to restore the in-band image blur produced by the anti-aliasing schemes.
Real-time adaptive-optics is a means for enhancing the resolution of ground based, optical telescopes beyond the limits previously imposed by the turbulent atmosphere. One approach for linear performance modeling of closed-loop adaptive-optics system involves calculating very large covariance matrices whose components can be represented by sums of Hankel transform based integrals. In this paper we investigate approximate matrix factorizations of discretizations of such integrals. Two different approximate factorizations based upon representations of the underlying Bessel function are given, the first using a series representation due to Ellerbroek and the second an integral representations. The factorizations enable fast methods for both computing and applying the covariance matrices. For example, in the case of an equally spaced grid, it is shown that applying the approximated covariance matrix to a vector can be accomplished using the derived integral-based factorization involving a 2D fast cosine transform and a 2D separable fast multiple method. The total work is then O(N log N) where N is the dimensions of the covariance matrix in contrast to the usual O(N2) matrix-vector multiplication complexity. Error bounds exist for the matrix factorizations. We provide some simple computations to illustrate the ideas developed in the paper.