Registered Data

[01136] Advances in Variational Models and PDEs for Images

  • Session Time & Room :
    • 01136 (1/3) : 4C (Aug.24, 13:20-15:00) @E819
    • 01136 (2/3) : 4D (Aug.24, 15:30-17:10) @E819
    • 01136 (3/3) : 4E (Aug.24, 17:40-19:20) @E819
  • Type : Proposal of Minisymposium
  • Abstract : Variational models and partial differential equations have been used to model various aspects of images, and this has led to many effective approaches to solve diverse image processing problems, such as image denoising, segmentation, reconstruction. This minisymposium will provide a venue for the latest advances in analysis and algorithm design for variational models and PDEs, and we will showcase state-of-the-art applications in image processing.
  • Organizer(s) : Gunay Dogan, Ronald Lok Min Lui
  • Classification : 68U10, 65D18, 62H35, 94A08, 68T45
  • Minisymposium Program :
    • 01136 (1/3) : 4C @E819 [Chair: Gunay Dogan]
      • [03732] Algorithms for Variational Segmentation of Regions and Boundaries
        • Format : Talk at Waseda University
        • Author(s) :
          • Gunay Dogan (National Institute of Standards and Technology)
        • Abstract : We propose several algorithms for variational segmentation of regions and boundaries in images. The algorithms come in Lagrangian and Eulerian flavors, and optimize variational models incorporating boundaries and region statistics, as well as various geometric regularizers. We demonstrate the advantages of each algorithm on several examples. Our algorithms are available in the open-source Python package scikit-shape.
      • [04275] Individual Tooth Segmentation in Human Teeth Images Using Pseudo Edge-Region Obtained by Deep Neural Networks
        • Format : Talk at Waseda University
        • Author(s) :
          • Chang-Ock Lee (KAIST)
          • Seongeun Kim (KAIST)
        • Abstract : In human teeth images taken outside the oral cavity with a general optical camera, it is difficult to segment individual tooth due to common obstacles such as weak edges, intensity inhomogeneities and strong light reflections. In this talk, we propose a method for segmenting individual tooth in human teeth images. The key to this method is to obtain pseudo edge-region using deep neural networks.
      • [05117] Joint solution of multi-task problems in imaging
        • Format : Talk at Waseda University
        • Author(s) :
          • Doga Gursoy (Argonne National Laboratory)
        • Abstract : Imaging in challenging conditions, such as with limited and uncertain data, typically involves solving multiple tasks, including image denoising, registration, segmentation, and various other reconstruction tasks. While the traditional approach has been to address these problems one at a time, solving them jointly with minimal manual hyperparameter setting can provide significant benefits in terms of image quality and acquisition time. In this presentation, I will explore the use of distributed optimization techniques to tackle these challenges and offer examples from my experience in the field of x-ray imaging.
      • [01799] Counting Objects by Diffused Index: geometry-free and training-free approach
        • Format : Talk at Waseda University
        • Author(s) :
          • Maryam Yashtini (Georgetown University)
        • Abstract : Counting objects is a fundamental but challenging problem. In this talk, I propose diffusion-based, geometry-free, and learning-free methodologies to count the number of objects in images. The main idea is to represent each object by a unique index value regardless of its intensity or size, and to simply count the number of index values. First, I place different vectors, referred to as seed vectors, uniformly throughout the mask image. The mask image has boundary information of the objects to be counted. Secondly, the seeds are diffused using an edge-weighted harmonic variational optimization model within each object. I propose an efficient algorithm based on an operator splitting approach and alternating direction minimization method, and theoretical analysis of this algorithm is given. An optimal solution of the model is obtained when the distributed seeds are completely diffused such that there is a unique intensity within each object, which I refer to as an index. For computational efficiency, I stop the diffusion process before a full convergence, and propose to cluster these diffused index values. I refer to this approach as Counting Objects by Diffused Index (CODI). I explore scalar and multi-dimensional seed vectors. For Scalar seeds, I use Gaussian fitting in histogram to count, while for vector seeds, I exploit a high-dimensional clustering method for the final step of counting via clustering. The proposed method is flexible even if the boundary of the object is not clear nor fully enclosed. I present counting results in various applications such as biological cells, agriculture, concert crowd, and transportation. Some comparisons with existing methods are presented.
    • 01136 (2/3) : 4D @E819 [Chair: Gunay Dogan]
      • [04373] A deep quasiconformal approach for topological preserving image segmentation
        • Format : Online Talk on Zoom
        • Author(s) :
          • Ronald Lok Ming LUI (The Chinese University of Hong Kong)
        • Abstract : In this talk, we address the problem of topology-preserving image segmentation based on quasiconformal (QC) theories. We introduce a variational model to obtain an optimal QC map that deforms a template mask to the segmentation mask while preserving the topology of the template mask. The bijectivity of the mapping is controlled by the Beltrami coefficient, which measures the QC distortion. We demonstrate that the proposed QC segmentation model can be effectively incorporated into a deep neural network architecture. The resulting deep QC segmentation network takes an image and a template mask with a prescribed topological prior as inputs and outputs the optimal QC map. The QC map is further used to deform the template mask to obtain the segmentation result. Experimental results show that the proposed approach outperforms existing state-of-the-art methods, making it a promising approach for topological preserving image segmentation. This work is supported by HKRGC GRF (Project IDs: 14306721,14307622).
      • [02523] Geodesic Models with Curvature Penalization for Image Analysis
        • Format : Talk at Waseda University
        • Author(s) :
          • Da CHEN (Shandong Artificial Intelligence Institute)
        • Abstract : Geodesic models establish the connection between the minimization of a weighted curve length and the viscosity solutions to the HJB PDEs. In contrast to globally minimizing a simplified first-order energy, as done by the classical geodesic models, we have recently extended the geodesic models to cover different curvature regularization terms, in conjunction with convexity shape prior and curvature prior constraint. We also show their applications in tubular structure tracking and image segmentation.
      • [02808] Texture edge detection via Patch consensus
        • Format : Talk at Waseda University
        • Author(s) :
          • Guangyu Cui (Georgia Institute of Technology)
          • Sung Ha Kang (Georgia Institute of Techonology)
        • Abstract : While well-known segmentation method are often based on homogeneity of regions, we focus on finding boundaries between different textured regions. We propose a training-free method to detect the boundary of texture by considering consensus of patch responses away from the boundary. We derive the necessary condition for textures to be distinguished, and analyze the size of the patch with respect to the scale of textures. Various experiments are presented to validate our model.
      • [04153] Density-equalizing map with applications
        • Format : Talk at Waseda University
        • Author(s) :
          • Gary Choi (The Chinese University of Hong Kong)
        • Abstract : We present surface and volumetric mapping methods based on a natural principle of density diffusion. Specifically, we start with a prescribed density distribution in a surface or volumetric domain, and then create shape deformations with different regions enlarged or shrunk based on the density gradient. By changing the density distribution, we can achieve different mappings including area-preserving parameterizations. Applications of the methods to medical shape analysis, data visualization, remeshing and shape morphing will be presented.
    • 01136 (3/3) : 4E @E819 [Chair: Gunay Dogan]
      • [01712] Application of weighted TV flow to material science problems
        • Format : Online Talk on Zoom
        • Author(s) :
          • Prashant Athavale (Clarkson University)
          • Emmanuel Atindama (Clarkson University)
          • Peter Lef (Clarkson University)
          • Gunay Dogan (National Institute of Standards and Technology)
        • Abstract : Several variational and partial differential equation (PDE)-based image processing methods can restore noisy crystallographic orientation data. We discuss restoration approaches, such as the classical total variation-based methods to diffusion PDEs. However, such methods are parameter-dependent, making them challenging in practice. Our work discusses an algorithm to restore noisy orientation data and circumvent the parameter selection problem by using weighted total variation flow, a nonlinear diffusion applied to the noisy orientation map.
      • [04202] Rank-One Prior: Real-Time Scene Recovery
        • Format : Online Talk on Zoom
        • Author(s) :
          • Tieyong Zeng (The Chinese University of Hong Kong)
        • Abstract : Scene recovery is a fundamental imaging task with several practical applications, including video surveillance and autonomous vehicles, etc. In this talk, we provide a new real-time scene recovery framework to restore degraded images under different weather/imaging conditions, such as underwater, sand dust and haze. A degraded image can actually be seen as a superimposition of a clear image with the same color imaging environment (underwater, sand or haze, etc.). Mathematically, we can introduce a rank-one matrix to characterize this phenomenon, i.e., rank-one prior (ROP). Using the prior, a direct method with the complexity is derived for real-time recovery. For general cases, we develop ROP to further improve the recovery performance. Comprehensive experiments of the scene recovery illustrate that our method outperforms competitively several state-of-the-art imaging methods.
      • [02233] Multispectral Image Restoration by Structured Eigendecomposition
        • Format : Talk at Waseda University
        • Author(s) :
          • Zhantao MA (The University of Hong Kong)
          • Michael Kwok-Po NG (The University of Hong Kong)
        • Abstract : We propose and study the opponent transformation for multispectral images. We generalize the well-known opponent transformation for color images and use it to bring the generalized opponent transformation total variation (GOTTV) multispectral image restoration model. By inheriting the crucial properties of the opponent transformation, the minimization formula of the GOTTV can be simplified and solved by the ADMM. Numerical examples are presented to demonstrate that the performance of the new GOTTV is well.