The Fourth International Workshop on Image Processing Techniques and Applications
22-23 July 2019
CMIT, University of Liverpool, UK
Image Processing Techniques and Applications (IPTA) is a Liverpool-based workshop for the communication of research related to image processing techniques and its applications to real life imaging problems. Held every 4 years, it provides an opportunity to present and discuss research on recent advances in mathematical modelling and analysis. A journal special issue is planned after the event in the Journal of Algorithms and Computational Technology (OA, but no charge).
Invited speakers include:
- Prof Raymond Chan, City University of Hong Kong, Hong Kong
- Prof Frederic Precioso, University of Nice-Sophie Antipolis, France
- Prof Fiorella Sgallari, University of Bologna, Italy
- Prof Bob Plemmons, Wake Forest University, USA
- Prof Laurent Cohen, Universite Paris Dauphine, France
- Prof Marta Betcke, University College London, UK
- Prof Dexing Kong, Zhejiang University, China
- Prof Joachim Weickert, Saarland University, Germany
- Prof Xiao-Chuan Cai, University of Colorado, USA
- Prof Rongji Lai, Rensselaer Polytechnic Institute, USA
- Prof Ronald Lui, Chinese University of Hong Kong, China
Aims and Themes
- review the state-of-the-art in image processing models, methods, techniques and commercial software packages;
- review learning methods in imaging,
- highlight new techniques and applications in Biomedical Imaging,
- tackle existing and emerging image challenges,
- promote mutual understanding of researchers in different disciplines, and networking;
- facilitate future research development and collaborations;
The main themes are image segmentation, multi-modality registration, deep learning and efficient algorithms with applications in medical and biomedical imaging.
IPTA 2019 is jointly sponsored by
- Centre for Mathematical Imaging Techniques (CMIT and LCMH), University of Liverpool
- Centre for Mathematical Imaging in Healthcare (CMIH), University of Cambridge
- Computer Vision and Machine Learning Lab (CVML), University of Central Lancashire
IPTA 2019 is organised by
- Ke Chen (Chair, CMIT/ LCMH and Math Sciences, University of Liverpool)
- Bogdan J. Matuszewski (Robotics and Computer Vision Lab, University of Central Lancashire)
- Harish Poptani (Centre for Preclinical Imaging, University of Liverpool)
- Carola-Bibiane Schönlieb (DAMTP, University of Cambridge)
- Yalin Zheng (Department of Eye and Vision Science, University of Liverpool)
Registration is open
The 4th International Workshop on Image Processing Techniques and Applications, is being held at the University of Liverpool on 22-23 July 2019, immediately before the 23rd Conference on Medical Image Understanding and Analysis (MIUA2019 on 24-26 July 2019).
The registration fee for this IPTA workshop is £50 but, if booked with MIUA, there is an overall discount of £25.
Liverpool, an amazing city, was named one of the top five places to visit in the UK by tourists from around the world. The venue is easily accessible via Manchester international airport (35 mins away), Liverpool John Lennon airport (20 mins away) and London Euston Train station (2 hours away).
Prof Raymond Chan, City University of Hong Kong, Hong Kong
We consider the high-resolution imaging problem of 3D point source image recovery from 2D data using a method based on point spread function (PSF) engineering. The method involves the use of a rotating PSF with a single lobe to obtain depth from defocus. The amount of rotation of the PSF encodes the depth position of the point source. Applications include high-resolution single molecule localization microscopy as well as the localization of space debris using a space-based telescope. Finding the locations and fluxes of the point sources is a large-scale sparse 3D inverse problem. A new nonconvex regularization method with a data-fitting term based on Kullback-Leibler divergence is proposed for 3D localization for the Poisson noise model. We will present numerical experiments to illustrate the efficiency and stability of the algorithm.
Joint authors: Chao Wang, Mila Nikolova, Robert Plemmons, and Sudhakar Prasad.
Prof Frederic Precioso, University of Nice-Sophie Antipolis, France
A representative selection of deep learning methodologies and their relevant applications on various medical problems will be presented.
Throughout this talk an emphasis is placed on recent theoretical results in deep learning. In addition, the role of these theoretical results on a range of applications will be demonstrated and explained. Application topics will include deep networks for interpretation of ECG data and use of deep learning techniques for processing and interpretation of multi-modal signals to analyse Surgical Workflow in the SensorOr (Endoscopic Vision Challenge). We will also look at how deep networks can support solutions for robust detection/prediction of adverse drug events using clinical notes. Lastly, we consider hardware architectures for deep learning implementations within a hospital environment to avoid computational outsourcing (e.g. cloud and web services).
Prof Fiorella Sgallari, University of Bolgna, Italy
A popular strategy for determining solutions to linear least-squares problems relies on using sparsity-promoting regularizers and is widely exploited in image processing applications such as, e.g., image denoising, deblurring and inpainting. It is well known that, in general, non-convex regularizers hold the potential for promoting sparsity more effectively than convex regularizers such as, e.g., those involving the l1 norm. To avoid the intrinsic difficulties related to non-convex optimization, the Convex Non-Convex (CNC) strategy has been proposed, which allows the use of non-convex regularization while maintaining convexity of the total objective function.
In this talk, a new CNC variational model is proposed, based on a more general parametric non-convex non-separable regularizer. The proposed model is applicable to a greater variety of image processing problems than prior CNC methods. Convexity conditions and related theoretical properties of the presented CNC model will be discussed. A primal-dual forward-backward splitting algorithm is proposed for solving the related saddle-point problem. Several numerical experiments are presented which prove the effectiveness of the proposed approach.
Joint work with Alessandro Lanza, Serena Morigi and Ivan Selesnick.
Prof Bob Plemmons, Wake Forest University, USA
Hyperspectral, Light Detection and Ranging (LiDAR), and Polarimetric imaging are the pervasive optical imaging modalities in remote sensing. Here, an overview is provided of our recent work on 3D optical imaging methods in rapid sensing for environmental as well as space safety concerns, along with test results. Topics overviewed include: (1) convolutional neural networks and spectral-spatial classification of (fused) hyperspectral and LiDAR data for urban and rain forest regions, (2) compressed sensing spectro-polarimetric imaging for space object material and shape information, and, especially, (3) joint 3D localization and classification of space debris using a multispectral rotating point spread function (RPSF). For (3) we show how the use of RPSFs allows one to estimate the 3D locations of point sources from a single 2D image. Going from a single wavelength to using multispectral images, with one RPSF per wavelength, we are able not only to localize the 3D positions of unresolved space debris but also classify their material composition.
Joint work, on topic (3), with C. Wang (UT-Dallas), S. Prasad (Minnesota) and G. Ballard (Wake Forest).
Prof Xiao-Chuan Cai, University of Colorado, USA
We discuss some highly scalable parallel domain decomposition algorithms for the simulation of blood flows in compliant human arteries by solving a system of nonlinear partial differential equations consisting of an elasticity equation for the artery and an incompressible Navier-Stokes system for the blood flow. The geometry of the artery is obtained from CT or MPI images. The differential equations are discretized with a fully implicit finite element method on unstructured moving meshes in 3D and solved by a Newton-Krylov algorithm preconditioned with an overlapping Schwarz method. Several mathematical, biomechanical, and supercomputing issues will be discussed in detail, and numerical experiments for the cerebral and pulmonary arteries will be presented. We will also report the parallel performance of the method on a supercomputer with a large number of processors.
Prof Ronald Lui, Chinese University of Hong Kong, China
We present a new framework to achieve image segmentation with convexity prior, guaranteeing the segmented region to be fully or partially convex. The basic idea is to incorporate a registration-based segmentation model with a specifically designed contour convexifier which is based on the discrete conformality structures of the image mesh. In the iteration process, while the segmentation model is deforming the given topological prior object to fit the color intensity, it is also convexified by the proposed contour convexifier. Ultimately, the target object is captured by a (fully or partially) convex region so as to minimize the effect of occluded or noisy boundary. Experiments have been carried out on both synthetic and real images and the results demonstrate the effectiveness of the proposed framework.
Dr Marta M Beckte, University College London, UK
In photoacoustic tomography, the acoustic propagation time across the specimen constitutes the ultimate limit on sequential sampling frequency. Furthermore, the state-of-the art PAT systems are still remote from realising this limit. Hence, for high resolution imaging problems, the acquisition of a complete set of data can be impractical or even not possible e.g. the underlying dynamics causes the object to evolve faster than measurements can be acquired. To mitigate this problem we revert to parallel data acquisition along with subsampling/compressed sensing techniques. Motivated by two results on near optimal sparsity of image representation and wave field propagation in Curvelet frame we consider methods for photoacoustic reconstruction under such sparsity assumptions in both image and data domain and discuss the relations between the two.
Prof Joachim Weickert, Saarland University, Germany
Partial differential equations (PDEs) of backward parabolic type have been advocated for image enhancement since more than five decades (Gabor 1965, Rudin/Osher 1991, Gilboa et al. 2002). Usually these models are supplemented with some stabilisation in order to tame their intrinsic ill-posedness. Nevertheless, most researchers refrain from using them because standard numerical schemes give rise to instabilities. In this talk we discuss a number of fairly general numerical tools to handle these challenging evolution equations. These include nonstandard discretisations and sequential splittings into highly localised processes, but also ideas from the numerics of hyperbolic PDEs such as upwinding, minmod schemes, and curvature limiters. Incorporating these concepts leads to theoretical stability guarantees which reflect key properties of the continuous models. Our experiments with forward-and-backward diffusion and evolutions of Gabor type demonstrate the practical usefulness and the good performance of the resulting algorithms.
Joint work with Martin Welk (UMIT, Hall, Austria) and Guy Gilboa (Technion, Haifa, Israel).
Prof Laurent Cohen, University Paris Dauphine, France
Active contours were introduced about 30 years ago as an interactive tool for object segmentation through energy minimization. Since active contours have the drawback of being trapped into local minima, we introduced about 20 years ago the minimal path method in order to find the global minimum of the active contour energy. Minimal paths have been used for long as an interactive tool to segment tubular structures as cost minimizing curves. The user usually provides start and end points on the image and gets the minimal path as output. These minimal paths correspond to minimal geodesics according to some adapted metric. They are a way to find a (set of) curve(s) globally minimizing the geodesic active contours energy. Finding a geodesic distance can be solved by the Eikonal equation using the fast and efficient Fast Marching method. Introduced first as a way to find the global minimum of a simplified active contour energy, we have recently extended these methods to cover all kinds of active contour energy terms. Through finding geodesics for new kinds of metrics, we were able to revisit active contour methods and propose the minimization of various active contours terms: Curvature penalization, Region term, Alignment term, as well as front propagation. We will present the mathematical background as well as concrete applications to biomedical and natural images.
Prof Rongii Lai, Rensselaer Polytechnic Institute, USA
In this talk, I will discuss our recent work of non-isometric matching for largely deformed shapes. Our method is based on Laplace-Beltrami basis pursuit via conformal deformation. My talk will further extend to introduce a new way of defining convolution on manifolds via parallel transport. This geometric way of defining convolution provides a natural combination of modeling and learning on manifolds. I will demonstrate its applications to shape matching using deep neural network based on parallel transportation convolution (PTC-net).
Prof Dexing Kong, Zhejiang University, China
Mathematical medicine is an interdisciplinary subject of mathematics, physics and medicine, which also involves computer science, information theory and large data science. Its purpose is not only to reconstruct the geometry of tissues, organs and focal areas, to report the relative position of different kinds of tissues, blood vessels, etc., and to give the quantitative description of various anatomical information; but also to predict the occurrence and evolution of various diseases, to characterize the occurrence mechanisms on diseases, to predict the treatment effect and survival prognosis, to reveal the inherent law of medical disciplines, and thus to help doctors to develop accurate personalized medical programs to achieve the ultimate goal for the benefit of each patient. This report will introduce some basic concepts, methods and theories of mathematical medicine, focusing on two clinical applications: precise radiofrequency ablation and medical image intelligent diagnosis system based on deep learning. These achievements have been successfully applied in the PLA General Hospital (Beijing 301 Hospital) and The First Affiliated Hospital of Zhejiang University.