2D, 3D and Surface Texture Analysis and Synthesis

Subscribers:
344,000
Published on ● Video Link: https://www.youtube.com/watch?v=Qc8MBAFK7x4



Duration: 57:13
616 views
4


Texture synthesis has been widely recognized as an important research topic. In this talk, I present a series of graphics or vision related projects that focus on texture analysis and synthesis. These projects cover 2D texture synthesis, 3D texture reconstruction, tensor-based BTF compression, static surface texture synthesis and dynamic surface flow simulation. Specifically, in 2D texture synthesis, we have introduced a new technique called feature-based synthesis which can produce superior results on aperiodic structural textures. It is based on the observation that the human visual system is most sensitive to edges, corners, and other high-level features in images. To produce high-fidelity texture results, we detect curvilinear features and explicitly match these features using a generalized distance metric including the Hausdorff distance. Regarding 3D texture analysis, we have developed methods for recovering the geometric details of 3D textures by exploiting shadows. We introduced a new concept called shadow graphs which give a novel graph-based representation for shadow constraints. Shadow graphs provide a much simpler and more systematic approach to represent and integrate shadow constraints from multiple images. To recover 3D textures from a sparse set of images, we developed an optimization method for integrating shadow and shading constraints. The rendered synthetic images of the recovered 3D textures are comparable to the original photographs. In terms of BTF compression, we developed an out-of-core tensor approximation algorithm. Traditionally, each multi-dimensional data item is represented as a vector. Such a scheme flattens the data and partially destroys the internal structures established throughout the multiple dimensions. In this work, we retain the original dimensionality of the data items to more effectively exploit existing spatial redundancy and allow more efficient computation. Experimental results indicate that our techniques can not only process out-of-core data, but also achieve higher compression ratios and quality than previous methods. Following the Poisson-based mesh editing technique jointly developed with MSR Asia, we have further applied the basic ideas to surface flow simulation and surface texture synthesis from multiple sources. I will briefly talk about our techniques and results on these topics. In particular, the surface flow project is for inviscid fluid simulation over triangle meshes. It can enforce incompressibility on closed surfaces by utilizing a discrete formulation of the Poisson equation. This is joint work with Narendra Ahuja, John Chang, Lin Shi, Hongcheng Wang, and Qing Wu.




Other Videos By Microsoft Research


2016-09-05Locality and Phases: Dynamic Structures of Large-Scale Program Behavior
2016-09-05The Physical and Mental States of Users Anytime and in Their Natural Environments
2016-09-05Skoll: Distributed Continuous Quality Assurance
2016-09-05Everything Bad Is Good for You: How Today's Popular Culture Is Actually Making Us Smarter
2016-09-05Service Placement in Stream-Based Overlay Networks
2016-09-05Protocol Composition Logics
2016-09-05THE FUTURE IS NOT FRAMED
2016-09-05Model Checking of Predicate Abstracted Programs without BDDs [1/2]
2016-09-05Social Computing Symposium - Panel - Extracting Signal from Noise in Social Networking
2016-09-05Menger's theorem for infinite graphs
2016-09-052D, 3D and Surface Texture Analysis and Synthesis
2016-09-05Anonymity in Peer-to-peer Systems
2016-09-05Patterns as Signs
2016-09-05Social Computing Symposium - Positive Externalities
2016-09-05Analyzing Mobile ad hoc Network Protocols via Probabilistic Model Checking [1/26]
2016-09-05Are aspects really needed for aspect-oriented programming?
2016-09-05Live Long and Prosper! Exercise, Nutrition and Supplements for Optimal Energy and Productivity
2016-09-05Social Computing Symposium - Visualizing Social Interactions and Collaboration History
2016-09-05Social Computing Symposium - Back Channels: Power and the Active Audience
2016-09-05Social Computing Symposium - Exploring the Social Institutional Dimensions of MoSoSo Design
2016-09-05A Search Engine for Natural Language Applications



Tags:
microsoft research