Browsing by Author "Alim, Usman Raza"

Now showing 1 - 4 of 4
  • Thumbnail Image
    ItemOpen Access
    Detecting Abnormalities in Thermal Pattern of Faces for Healthcare Applications
    (2019-05-14) Ejindu, Oluchukwu Roseline; Yanushkevich, Svetlana N.; Nowicki, Edwin Peter; Alim, Usman Raza
    In this work, we propose a novel method of applying deep learning technique in thermal image processing and analysis for healthcare application. It addresses detection of abnormal thermal patterns, thus identifying, in particular, patterns of elevated temperature that indicate fever, hypothermia and related abnormalities. Temperature estimation is performed based on the analysis of regions-of-interest from the thermal images of human faces. Another focus of this work is to investigate thermal effects of alcohol intoxication. We applied the deep learning approach on 16,000 usable images of 40 subjects from a publicly-available Drunk-Sober database. Two Convolutional Neural Network architectures were investigated for the task of processing of two regions of interest - the forehead and the eyes. The accuracy of the neural network classifiers to predict subject’s insobriety using the forehead and eye regions-of-interest reached 95.5% and 96.67%, respectively, compared to the best-known results on the same data using a non-deep neural networks. To boost the accuracy of classification, both the feature-level and the score-level fusion were applied as well, thus improving the accuracy to 96.92%.
  • Thumbnail Image
    ItemOpen Access
    Spatial Partitioning for Distributed Path-Tracing Workloads
    (2018-09-21) Hornbeck, Haysn; Alim, Usman Raza; Gavrilova, Marina L.; Chan, Sonny
    The literature on path tracing has rarely explored distributing workload using distinct spatial partitions. This thesis corrects that by describing seven algorithms which use Voronoi cells to partition scene data. They were tested by simulating their performance with real-world data, and fitting the results to a model of how such partitions should behave. Analysis shows that image-centric partitioning outperforms other algorithms, with a few exceptions, and restricting Voronoi centroid movement leads to more efficient algorithms. The restricted algorithms also demonstrate excellent scaling properties. Potential refinements are discussed, such as voxelization and locality, but the tested algorithms are worth further exploration. The details of an implementation are outlined, as well.
  • Thumbnail Image
    ItemOpen Access
    Visual Analytics Framework for Exploring Uncertainty in Reservoir Models
    (2018-08-31) Sahaf, Zahra; Costa Sousa, Mário; Maurer, Frank; Willett, Wesley; Alim, Usman Raza; El-Sheimy, Naser; Mackay, Eric James
    Uncertainty is related to poor knowledge of a phenomenon. In particular, geological uncertainty is an essential element that affects the prediction of hydrocarbon production. The standard approach to address the geological uncertainty is to generate a large number of random 3D geological models and then perform flow simulations for each of them. Such a bruteforce approach is not efficient as the flow simulations are computationally costly and as a result, domain experts cannot afford running a large number of simulations. Therefore, it is critically important to be able to address the uncertainty using a few geological models, which can reasonably represent the overall uncertainty of the ensemble. Our goal is to design and develop a visual analytics framework to filter the geological models and to only select models that can potentially cover the uncertain space. In this framework, a new block based approach is proposed using mutual information to calculate pair-wise distances between the 3D geological models. The calculated distances are then used within a clustering algorithm to group similar models. Cluster centers are the few representative models of the entire set of models that cover the uncertainty range. The whole framework is complimented by visual interactive tasks to be able to incorporate user's knowledge into the process and make the entire process more understandable. Finally, the framework is applied on many different case studies, and the results are evaluated by comparing with the existent brute force approach. In addition to that, the actual framework is evaluated in formal user study sessions with the domain experts in reservoir engineering and geoscience domain.
  • Thumbnail Image
    ItemOpen Access
    Visualization of Multivariate Data on Surfaces
    (2019-03-19) Rocha, Allan; Costa Sousa, Mario; Alim, Usman Raza; Chan, Sonny; Jacob, Christian J.; Geiger, Sebastian; Tominski, Christian
    In several domains of science and applications, the understanding of scientific data leads to technological advances and scientific discovery. Multivariate 3D data, for example, is essential for decision-making in fields such as Medicine and Geology, where experts are required to understand and correlate several spatial attributes. To simplify complexity and facilitate understanding, the 3D data is often explored through surfaces of interest. This is the reason why the visualization of multivariate data on surfaces has been a topic of interest among the visualization community. However, much work has been needed to provide visualization solutions that facilitate the multivariate visualization design, creation, and exploration. This research builds upon ideas introduced and discussed many years ago that focus on the problem of visualizing multiple attributes on surfaces in a single view. Here I present a new perspective to this problem as well as a solution that allows us to design, visualize and interact with multivariate data on surfaces. This perspective is created from the combination of several aspects born in fields such as Illustration, Perception, and Design, that have been employed and studied by the visualization community both in Information and Scientific Visualization. Therefore, this thesis lies between these two main fields, since it involves aspects from both. By building upon this multidisciplinary combination, I present a new way to visualize multivariate data on surfaces by exploiting the concept of layering. First, I introduce a new real-time rendering technique and the concept of Decal-Maps, which fills a gap in the literature and allow us to create 2D visual representations such as glyphs that follow the surface geometry. Building on this technique, I propose the layering framework to facilitate the multivariate visualization design on surfaces. The use of this concept and framework allows us to connect and generalize concepts established in flat space, such as 2D maps, to arbitrary surfaces. This thesis also demonstrates that the design of new multivariate visualizations on surfaces opens up other new possibilities such as the use of interaction techniques. Here I demonstrate this potential by introducing a new interaction technique that allows us to explore multivariate data and to create customized focus+context visualizations on surfaces. This is achieved by introducing a new category of lenses, Decal-Lenses, which extends the concept of magic-lenses from flat space to general surfaces. Finally, this thesis showcases the process of multivariate visual design and data exploration through a series of examples from several domains. Inspired by these examples, I also contribute with an in-depth application research conducted from my long-term collaboration with domain experts in the fields of Geology and Reservoir Engineering. This application illustrates how the proposed approach can support and facilitate decision-making in the complex process of Geological Modelling.