Sanjay Tiwari

Sanjay Tiwari is studying the ocean floor and its features to identify commercially important species and their preferred habitats. Yet Tiwari, a research specialist in the WHOI biology department, is leading the Sea Grant-supported project from his desk, not a ship. Tiwari is a mathematician and he is using math to classify benthic habitats in a project that could play a significant role in fisheries management. Over the years, a variety of techniques and tools have been developed to help with habitat identification: side-scan sonar, remotely operated vehicles, video imaging, and submersibles. Tiwari says that while each has important capabilities, collectively they share "an inability to characterize images [in a way that] describes substrate and habitat automatically and rapidly." What's needed, he feels, "is a set of automated image processing tools that can classify an infinite variety of habitats based on a scheme that we, as humans, consider important to benthic organisms." But first, he notes, those habitats must be defined and characterized.

NOAA's Northeast Region Essential Fish Habitat Steering Committee took on part of the task a few years ago, creating eight benthic habitat categories. Tiwari is working on another key step: looking at biological structures to determine what living organisms are attached to, buried in, or resting on each background. And he's using digital images - from video or fast still camera - to supply the data. The problem? Each image contains vast amounts of data that must be processed, analyzed, and characterized.

Enter Tiwari, with his array of mathematical approaches programmed to interpret the images. One, wavelet analysis, is a way of compressing an image using scale (resolution) and frequency (contrast) to represent the original image in a more efficient way. JPEG2000, a popular image format, is an example of using wavelets to store images.

Once images have been compressed, Tiwari creates and applies "learning machines," or algorithms, to distinguish patterns in the images. Learning machines differ in the ways in which they compute "similarity" among patterns. Tiwari can identify a scallop by assigning it a set of numbers that correspond to its characteristics or "feature sets," then using learning machines to distinguish it from other feature sets, say those of clams and sand dollars.

Tiwari is now working to increase the sophistication of the algorithms. And he's applying another mathematical approach called the hidden Markov model (HMM), commonly used in multiple DNA sequence alignment and gene finding.

Math and benthic species are just a couple examples of Tiwari's diverse interests. He was born in the foothills of the Himalayas. Tiwari's father was an economist, which may explain his affinity for numbers. His mother studied English literature, which likely led to Tiwari's being "hooked on Victorian and Edwardian literature." He collects first editions of Victorian and Edwardian children's books. And the hobby has led to a writing project of his own: Tiwari is now rewriting Wikipedia's Rudyard Kipling page. Oh, and he's also a runner and blues harp player - though he hasn't figured out a way to combine those interests...yet.

- Tracey Crago, Woods Hole Sea Grant