/ Karol Myszkowski / Associate Professor
/ A. G. Belyaev / Associate Professor
/ Ilia A. Bogaevski / Visiting Researcher
/ Runhe Huang / Assistant Professor
/ Przemyslaw Rokita / Visiting Researcher
The research conducted in the Computer Science and Engineering Laboratory is currently concentrated on the following projects:
1. Realistic Image Synthesis
The main goal of the project is enhancement of realism in virtual environments via careful simulation of lighting, and seamless real-time rendering. The research is focused on the following topics:
Human perception-based metrics of the image quality, and perception-driven lighting simulation algorithms. Many improvements in the computational efficiency of rendering can be attained by focusing computations only on those scene features readily perceived by human observers under given viewing conditions.
Lighting simulation using stochastic photon tracing and density estimation techniques. All visually important lighting phenomena can be easily simulated within this framework, however, efficient techniques of lighting storage and reconstruction are of key importance to make this technique suitable for practical applications.
Image-based rendering and virtual walkthroughs in the real-world environments. Photographs are very effective and inexpensive way to acquire information about the existing real-world objects. In this research, techniques for deriving images for arbitrary camera positions based on a finite number of photographs are investigated.
2. Differential Equations, Differential Geometry, and Singularity Theory for Image Processing and Geometric Modeling
In recent years it has been significant progress in applications of sophisticated mathematical theories to problems arising in image processing and computer graphics. Our current research is focused on methods based on partial differential equations, classical differential geometry, and singularity theory. It includes curvature-based shape and image analysis, singularity theory for shape interrogation, and homogenization of partial differential equations in media with periodic structures.
3. Designs of a Global Teaching and Learning aimed EducationalSystem
Recent years, advances and great progresses in multimedia computing, network and the Internet techniques have brought about an educational revolution, i.e., {\it global teaching and learning}, which means that teachers and students can conduct their teaching/learning activities in anytime at anywhere so long as they have computers connected to the Internet. To enable global teaching and learning, a networked computer based education system is proposed and under developing.
The educational system consists of five main components: virtual office, virtual classroom, virtual collaborative room, virtual laboratory, and virtual library. The virtual classroom is a system to support preparation and authoring of teaching materials, to effectively organize teaching activities, and to ease students learning activities. Supplement environment, evaluation environment, exploring environment, and collaborative environment are developed and integrated as a hyper-environment to support the virtual classroom. The virtual collaborative room is a system to provide good interactions and communications among collaboration members. Different types of rooms are defined and constructed to support different types of collaborative activities. Framework of intelligent agent system is established to support automation of teaching and learning activities in an integrated education system.
Refereed Journal Papers
We describe singularities of the convex hull of a generic compact smooth hypersurface in four-dimensional affine space up to diffeomorphisms. It turns out the boundary of the convex hull is the front of a Legendre variety. Its singularities are classified up to contact diffeomorphisms.
A viscosity solution of a Hamilton--Jacobi equation is the asymptotics of the solution with the same initial condition of the original Hamilton--Jacobi equation regularized by vanishing viscosity. Even if the initial condition is smooth, the viscosity solution can have singularities. In the case of a convex smooth Hamiltonian and a generic smooth initial condition we describe a full classification of these singularities and their perestroikas (= bifurcations, metamorphoses) in spaces of physically interesting dimensions 1, 2, and 3.
In this paper, we discuss efficient techniques for storage and rendering of high-quality lighting in interactive applications which deal with complex, static scenes, and require frequent changes of viewing parameters.
In this paper, we consider accelerated rendering of walkthrough animation sequences using combination of ray tracing and Image-Based Rendering (IBR) techniques. Our goal is to derive as many pixels as possible using inexpensive IBR techniques without affecting the animation quality. A perception-based spatio-temporal Animation Quality Metric (AQM) is used to automatically guide such a hybrid rendering. The Pixel Flow (PF) obtained as a by-product of the IBR computation is an integral part of the AQM. The final animation quality is enhanced by an efficient spatio-temporal antialiasing, which utilize the PF to perform a motion-compensated filtering.
In this study of global illumination computations, we investigate the applications of the perceptually-based Visual Difference Predictor (VDP) developed by Daly. First, we validate the performance of this predictor in shadow masking by texture and luminance contrast experiments. We also experiment with Contrast Sensitivity Functions (CSFs) derived from the results of various psychophysical experiments, various spatial frequency and orientation channel decomposition schemes, and contrast definitions, in order to check predictor integrity and sensitivity to differing models of visual mechanisms. We show applications of the VDP to monitor the perceived quality of the progressive radiosity and Monte Carlo solutions, and decide upon their stopping conditions. Also, based on the local error metric provided by the predictor we show some initial attempts to drive adaptive mesh subdivision in radiosity computations.
Current thought on global illumination solutions is that they require long hours of computation for complex scenes, and because of that they are mostly used only at the final stages of the process of scene design and rendering. This may result in severe obstruction of the design process because of the substantial differences in appearance that can occur between the final image and the intermediate images, the latter usually being based upon simplistic lighting computations. This article presents a practical, view-independent, progressive global illumination technique, which takes into account some basic properties of human visual perception to provide the high quality images of complex environments within single minutes or seconds using physically-based partial solutions.
In order to fully develop and optimize this technique, a perceptually-informed framework was incorporated into the computational approach as a means to evaluate progressive changes in image quality. Instead of focusing on one particular algorithm, a pool of complementary algorithms was chosen, and the best technique was selected at every stage of the computations so that perceived differences between the intermediate and final images are minimized over time. The perception-based Visible Differences Predictor (VDP) developed by Daly \cite{Daly93} was used to obtain quantitative measures of such differences, and to support off-line decisions regarding the points in the computational process at which one algorithm from the pool should be switched with another. The pool contains well-known radiosity and Monte Carlo Photon Tracing (MCPT) techniques, which were composed in non-standard way, and enhanced to improve their performance. In particular, the complexity of the mesh used to store and display lighting was reduced by utilizing local estimates of the global illumination solution. The novel feature here was the availability of such local estimates at the early stages of computation to guide the adaptive mesh subdivision. Also, an efficient object space filtering, which substantially reduced perceivable noise inherent to stochastic solutions was proposed, and an inexpensive estimate of the convergence error for such solutions was provided.
The perceptually-based Visual Difference Predictor (VDP) developed by Daly \cite{Daly93} has many potential applications in realistic image synthesis. However, systematic validation and subsequent calibration of the VDP response via human psychophysical experiments should be completed before integrating the VDP into image synthesis algorithms such as those in global illumination computations. For example, the VDP local error metric can guide decision making in adaptive mesh subdivision, and in selecting regions of interest for more intensive global illumination computations. In this study, we designed two human psychophysical experiments to test whether VDP predictions match well with subjective reports of visible difference between images under conditions mimicking those in our VDP applications. These experiments showed a good match with VDP predictions for shadow and lighting pattern masking by texture, and in comparisons of the perceived quality of images generated at subsequent stages of indirect lighting solutions.