Tensor Displays are a family of light field displays. This is a little different from Layered 3D, for instance, which is bound to one particular hardware configuration – multiple layers. Tensor Displays can also be built from multiple layers of transparencies or LCDs (only in attenuating mode, as opposed to polarization-rotation mode) but also from any combination of lenses and layers. One more characteristic feature of most tensor displays is that they take full advantage of high-speed LCD panels. The human visual system (HVS) is rather slow and can, depending on the individual and the lighting conditions, only resolve up 60 frames per second. Showing time-varying patterns on a 120 Hz or 240 Hz LCD panel will result in the user perceiving an average over multiple patterns. What really defines the Tensor Display family is the mathematical framework they all share for driving any of the possible hardware implementations – non-negative light field tensor factorizations. These factorizations automatically compute optimal, time-varying pixels states for a given light field. Tensor Displays support significantly wider fields of view, depths of field as well as 3D image quality compared to Layered 3D and Polarization Fields. We explain all the mathematical details in our SIGGRAPH 2012 paper, which was featured as a SIGGRAPH 2012 Paper Highlight, an Emerging Technologies Highlight, and widely covered in the popular press including New Scientist, The Boston Globe, engadget, slashdot, and many more.
This is a conceptual illustration of Tensor Displays. You and your friends can enjoy glasses-free 3D content within a wide viewing zone. The Tensor Display itself is composed of some combination of high-speed LCDs and direction backlighting (which can for instance be implemented with lenslet arrays).
Here is one incarnation of a Tensor Display we built. The hardware configuration is related to Layered 3D and Polarization Fields, but the improved tensor factorizations can take full advantage of the speed of the LCDs and facilitate wider viewing zones and larger depths of field.
While most layered hardware configurations require a bit of depth, this Tensor Display is thin due to the directional backlight that it uses. While we hope to get our hand on a high-quality directional backlight one day, we improvised by emulating it with two LCDs that have a lenslet array sandwiched in between them.
Here is an overview of the construction. Follow the instructions to build your base monitor, then follow more detailed instructions on getting one or two additional LCDs; the above images just show an overview of the entire procedure. Once you have the components in place, you can simply stack up three LCDs or two LCDs with a lenslet array in between.
We have real-time OpenGL and offline Matlab code available for calibrating and computing patterns for the multilayer Tensor Display version. Due to some legal issues, we can currently not release the full tensor factorization framework that supports high-speed panels and the directional backlight. We hope you still enjoy your multilayered Tensor Display!