9 results for Andrews, Mark

  • Abundance Guided Endmember Selection: An Algorithm for Unmixing Hyperspectral Data

    Dowler, S; Andrews, Mark (2010)

    Conference item
    The University of Auckland Library

    Linear unmixing is a blind source separation problem that decomposes a hyperspectral image into the spectra of the material constituents of the scene and the abundance maps of those materials across that scene. A novel method for determining the material spectra from within the scene, AGES, is proposed based on the positional information contained within abundances generated by additivity-constrained inversion. This new approach is compared on both simulated and real data sets to the well established N-FINDR algorithm, comparing favorably in terms of computational complexity with the existing algorithm without significantly sacrificing accuracy. In addition, the algorithm has some desirable properties inherent in such an approach.

    View record details
  • Construction of a practical hyperspectral image acquisition system

    Bowmaker, R; Dunn, RJ; Moynihan, KB; Roper, TJ; Andrews, Mark (2011-11)

    Conference item
    The University of Auckland Library

    Abstract—A hyperspectral image acquisition system has been constructed to collect data in the visible and near-infrared wavelength bands and to support research in imaging spectroscopy. The design solves a number of practical difficulties associated with hyperspectral imaging related to workflow, exposure budget, reflectance calibration and, most importantly, chromatic aberration. Chromatic aberration, present in nearly all nonmonochromatic images, is particularly severe in hyperspectral images and has been significantly reduced by tracking the focal length of the lens as the wavelength changes. Attendant problems such as lens growth during focussing have also been addressed. The system is tightly integrated with software and streamlines an otherwise laborious and potentially error-prone capture process. The system is shown to provide higher quality images (as measured by the modulation transfer function) in an efficient manner with minimal chromatic aberration, simplified workflow and efficient exposure times.

    View record details
  • Compressively Sensed Hyperspectral Image Recovery using Total Variation Minimisation by Approximation

    Eason, D; Lee, William; Andrews, Mark (2012-12)

    Conference item
    The University of Auckland Library

    Compressive sensing (CS) theory states that any signal that is sparse in a known basis may be recovered from a small set of linear combinations of the signal. Capturing data a fraction of the size of the original signal may be beneficial when applied to hyperspectral imaging (HSI), an imaging technique that introduces spectral content in order to classify materials in a scene. Recently, total variation (TV) minimisation algorithms have achieved success in recovering images captured using CS. In this paper we present a novel implementation of the TV minimisation algorithm that includes a differentiable approximation of the TV norm. This new method compares favourably with other TV minimisation algorithms, and we show how it can be extended from monochromatic to hyperspectral CS image recovery.

    View record details
  • Compressive Hyperspectral Image Sensing Restoration via Joint TV-L1 Regularisation

    Lee, William; Eason, D; Andrews, Mark (2012-12)

    Conference item
    The University of Auckland Library

    Compressive sensing is a recently introduced signal acquisition technique that can significantly reduce the number of measurements necessary for reconstructing a signal to a high degree of accuracy. This has considerable practical benefits to hyperspectral imaging where the current cost and effort required to collect the vast amount of data hinders its applicability to numerous field of sciences. Current state-of-the-art algorithms for compressive sensing are designed for solving problems with single regularisers (such as the sparse-inducing L1-norm and total variation). This paper presents a method for solving compressive hyperspectral image sensing problem using joint L1-norm and total variation regularisation as an on-going effort in improving compressive hyperspectral image restoration. Preliminary results suggest that such regularisation gives better results than using the individual regularisers alone. Current work in progress includes incorporating spatial-spectral regularisation for hyperspectral image recovery and analysing new compressive sampling schemes to exploit such regularisation.

    View record details
  • Subspace estimates for real-time hyperspectral video systems

    Andrews, Mark; Dunn, R (2012-02-09)

    Journal article
    The University of Auckland Library

    A method is presented for rapidly estimating the principal components used in dimension reduction for high frame rate hyperspectral video systems. Pixels entering and leaving a video frame are used to rapidly update the covariance matrix (Σ) and estimate the perturbed eigenpairs. Measuring the angle between the estimated dominant eigenspace and the exact eigenspace shows the method to be a good approximation when the difference between frames is low (||δΣ|| ≪ ||Σ||). This method reduces the time to calculate the principal components of high resolution hyperspectral video data by a factor of ~20.

    View record details
  • Iterative blind deconvolution of extended objects

    Biggs, David S.C.; Andrews, Mark (1997)

    Conference paper
    The University of Auckland Library

    An open access copy of this article is available and complies with the copyright holder/publisher conditions. This paper describes a technique for the blind deconvolution of extended objects such as the Hubble Space Telescope (HST), scanning electron and 3D fluorescence microscope images. The blind deconvolution mechanism is based on the Richardson-Lucy (1972, 1974) algorithm and alternates between deconvolution of the image and point spread function (PSF). This form of iterative blind deconvolution differs from that typically employed in that multiple PSF iterations are performed after each image iteration. The initial estimate for the PSF is the autocorrelation of the blurred image and the edges of the image are windowed to minimise wrap around artifacts. Acceleration techniques are employed to speed restoration and results from real HST, electron microscope and 3D fluorescence images are presented

    View record details
  • High resolution image reconstruction using mean field annealing

    Numnonda, Thanachart; Andrews, Mark (1994)

    Conference paper
    The University of Auckland Library

    An open access copy of this article is available and complies with the copyright holder/publisher conditions. A high resolution image can be reconstructed from a sequence of lower resolution frames of the same scene where each frame taken by the camera is offset by a subpixel displacement. In this paper, it is shown that such a reconstruction task can be cast as an optimisation problem, and that a reconstruction can be found using the mean field annealing algorithm. The proposed technique has the added advantage over existing techniques of not requiring the registration of the displacement of each low resolution frame. In addition, the proposed technique greatly reduces the required computation as compared to a simulated annealing approach

    View record details
  • Hyperspectral Video for Forensic Applications

    Andrews, Mark; Dunn, RJ (2011-11)

    Conference item
    The University of Auckland Library

    Hyperspectral Imaging is a technique where material properties and quantities are determined using interactions between light and matter. Its non-contact and non-destructive nature ensures it has numerous applications in forensic science. Many of these applications require a video-rate hyperspectral system, although, processing this volume of data is demanding and difficult to perform in real-time. A novel method is presented for reducing the complexity of current hyperspectral imaging techniques in the context of a hyperspectral video crime scene analysis tool. Specifically, the essential but time-consuming phase of dimension reduction is achieved using a new on-line estimate of the principal components and exploits temporal redundancy in sequential hyperspectral volumes. This new algorithm is shown to provide a significant reduction in complexity (> 10× ) in processing hyperspectral video when coupled with Abundance Guided Endmember Selection—a new endmember identification and extraction algorithm developed for hyperspectral video applications. A theoretical frame-rate of over 20fps for a scene with 5×10⁶ pixels and 224 bands can be achieved when implemented on an nVidia Tesla C2070.

    View record details
  • On the Convergence of N-FINDR and Related Algorithms: To Iterate or Not to Iterate?

    Dowler, S; Andrews, Mark (2010-12)

    Journal article
    The University of Auckland Library

    A popular algorithm for unmixing hyperspectral data, namely, Winter's N-FINDR algorithm, is frequently used to benchmark other algorithms or as the basis for new algorithms. The interpretations of this algorithm within the literature are not consistent, and some of these differences have significant impact on the convergence of the algorithm. Despite this, the differences in implementation have not been explicitly acknowledged within the literature, which means that many studies are now ambiguous or incomparable. An examination of various implementations of the N-FINDR algorithm highlights that not all interpretations possess the properties asserted by Winter and that interpretations that consider each pixel multiple times generate much larger simplexes. Regardless of which implementation researchers choose to use, if they are explicit in their choice, this would allow for unambiguous comparisons.

    View record details