Josh Smith

Joshua R. Smith

I finished my PhD in 1999, under Neil Gershenfeld. The title of my dissertation is Electric Field Imaging, "seeing" with electric fields. Electric Field Imaging is a new channel for machine perception that has been overlooked, perhaps because humans do not sense electric fields. I developed signal processing techniques, hardware, and inference algorithms to let machines acquire useful geometrical information from electric field measurements.

The School of Fish, developed for Electric Field Imaging, is a network of intelligent electrodes, each with its own analog hardware and on-board computing, on a shared communication channel. It is an unusual form of Sensor Network, because the electrodes are not actually capable of sensing individually. Sensing only occurs via pairwise interactions between electrodes, which makes sensing a property of the network itself, not of individual units.

Elesys (formerly NEC Automative Electronics) has released a product based directly on work I did in the course of my thesis. The product is a car seat with embedded electric field sensors to determine the size and body configuration of the occupant in order to make more intelligent firing decisions. The Occupant Position Detection System now ships in all Honda cars with side airbags, and an even more sophisticated Electric Field Imaging system from Elesys will be in GM vehicles beginning in Model Year 2004. Here is a televised demonstration of the Occupant Position Detection System.

Motorola has also launched an Electric Field Imaging IC, the MC33794 Electric Field Imaging Device.

The fact that Electric Field Imaging has so successfully addressed the important problem of automotive passenger sensing is, in my view, proof that alternative sensing mechanisms can enable better machine perceptual systems.

While still working on Electric Field Imaging, I realized that the weak signal detection techniques I was using could be applied in a very different setting, digital watermarking or data hiding. My paper Modulation and Information Hiding in Images was one of the first to propose a quantitative model of digital watermarking, and to frame digital watermarking in terms of communications concepts such as signal, noise, bandwidth and jamming margin. Here is a link to citations.

After finishing my PhD, I became Director of Escher Labs, which developed sensing, signal processing, and security technologies for pervasive information processing, with an emphasis on adding intelligence to paper documents. We developed a technology called FiberFingerprint that takes the next step beyond digital watermarking, enabling copy protection for physical media. Here is link to The Document That Can't be Forged, a New York Times article about FiberFingerprint (here is a local copy in case you have trouble logging in). Every square centimeter of paper has a unique pattern of hills and valleys. With the proper sensing and signal processing, we can make use of those characteristics for identification and authentication. FiberFingerprint also makes use of the weak signal detection principles used in Electric Field Imaging and data hiding.

Next I joined Tiax (formerly Arthur D. Little's Technology and Innovation Practice).

I am now a principal investigator at Intel Research Seattle.