PA1: Contributed Session

Room: Old Main Academic Center 3030

Webex Link

Chair:  Seongjai Kim, Mississippi State University

Seongjai Kim, Mississippi State University

Time: 4:00 pm - 4:20 pm (CST)

Title: Iterative Update Sample Consensus (IUSAC): A Repeatable Algorithm for Optimal Consensus Set


The RANdom SAmple Consensus (RANSAC) is one of the powerful tools for the reconstruction of ground structures from point cloud observations in many applications. The algorithm utilizes iterative search techniques for a set of inliers to find a proper model for the given data. Despite its successful performances in many fields, the RANSAC is not repeatable, i.e, the optimal solution is not guaranteed by its repeated run of the RANSAC. In this article, we propose a new algorithm for the reconstruction, called the Iterative Update SAmple Consensus (IUSAC), which is repeatable to get the optimal estimation for the model parameters. The main idea behind the IUSAC is the iterative update, which makes the estimated parameters converge to the optimal solution. The proposed algorithm is analyzed for convergence. It has been numerically verified that the IUSAC is repeatable, simple, and more efficient than the RANSAC. Several numerical results including the KITTI data validate the performance of the IUSAC. The KITTI data is a public domain dataset to serve for R&D activities on autonomous vehicle tracking in GPS-denied environments.

(joint work with Jichul Kim, Byungjoon Lee, Hwan Hee Park, Michael R. Zanetti, and Kyle A. Miller)

Zengyan Zhang,  Utah State University

Time: 4:20 pm - 4:40 pm (CST)

Title: Structure-preserving Reduced-order Models for Reversible-Irreversible Thermodynamically Consistent PDEs


As a powerful approach for dimensionality reduction, the proper orthogonal decomposition reduced-order model (POD-ROM) has been widely used as a computationally efficient surrogate model for complex large-scale systems. Given the computational complexity of the thermodynamically consistent models, the POR-ROM plays an important role in reducing the spatial-temporal complexity. However, the classical POD-ROM can destroy the thermodynamic structure in the reduced-order modeling approach for the systems. So far, designing structure-preserving reduced-order models for Hamiltonian systems (reversible) has been widely studied while there is little work on dissipative (irreversible) systems, i.e., the free energy for the system is dissipative while the temperature fluctuation is ignored. In this talk, we will introduce a numerical platform that can systematically derive ROMs for reversible-irreversible thermodynamically consistent PDEs by embracing several novel techniques and demonstrate its effectiveness in several numerical examples.

Victor BOGDAN, West University of Timisoara

Time: 4:40 pm - 5:00 pm (CST)

Title: An image sharpening technique based on dilated filters and 2D-DWT image fusion


Image sharpening techniques are common in the field of Image Processing, and they are used to increase the contrast between dark and bright regions in images. By using these techniques one can obtain images which contain finer and more prominent features. Previous research in the field suggested that by dilating kernels bring benefit effects to edge detection algorithms [1] yielded in some of the use cases better performances. Inspired by the positive sharpening results by multi-scale Unsharp Masking (UM) [2] technique scheme and a wavelet based fusion algorithm presented in [3], we present a novel multi-level dilatation wavelet scheme. The proposed sharpening algorithm processes the input image by a cluster of UM with different dilatation factors followed by a wavelet fusion. Visual results are presented in Figure 1 and we observe that the proposed algorithm bring clear enhancement of the image at hand, without any crisping effect. In order to highlight the enhancements we present the difference between images. The obtain results were measured using BRISQUE [4], a no-reference image quality assessment, and an improvement was observed in the metric from 0.53 for the original image to 0.79 for the proposed scheme.

This talk is based on a joint work with Ciprian ORHEI, and Cosmin BONCHIS.