The nation’s youth is thinking about a semantic web. At least some of them are. On the 2010 list of Intel Science Talent Search 2010’s 40 Finalists is an entry dubbed “Semantic Image Retrieval and Interactive Exploration of Large Image Collections,” the brainchild of 18-year old David Chienyun Liu, a student at San Jose’s Lynbrook High School.

The synopsis of the project is described as this: What began as an effort to organize the family photo album morphed into “software that searches images automatically by linking visual cues with concepts (called semantic concept models), such as trees and buildings, to improve retrieval precision by 22%. Then, using a spring graph, he created an interactive exploration system that can effectively group conceptually similar images. Working from home, he tested the system by examining aerial images that he obtained from NASA to identify hazards to buried oil pipelines and achieved a 96% recognition rate.”

Liu's bio on the Society for Science and the Public web site, also notes that Liu--co-president of his school’s robotics team and founder of the computer club--thinks this technology can be used in real time by unmanned aerial vehicles to safeguard pipelines and other valuable resources. It also could have applications for medical imaging and, of course, web search engines.

Perfecting image search is the focus of efforts by big names in the web search engine space, of course, as it's a challenge given that keywords for image search are often incomplete or inaccurate. Google, for example, has in beta Image Swirl, which clusters similar images into algortihmically-determined representative groups, and it currently works for more than 200,000 queries.

Perhaps a lesser-known effort on the semantic image search front are projects such as Pixolu, a prototype Internet image search system using the pixolution technology that was developed as a student project by Simon Burkard, Dominik Deichsel, Jörn Ehmann, Friedrich Maiwald and Ben Suksanguan under the supervision of Prof. Dr. Kai Uwe Barthel. The search system is based on three techniques: a visual sorting of images on a local computer that tries to group large sets of images by that which is similar to them (e.g. pictures of the Eiffel Tower at night, pictures of the Eiffel Tower by day, charcoal drawings of the Eiffel Tower, and so on); visual similarity search, where particular images can be used as a template to find more images like them; and suggestions of semantically similar images based on tracking what users have been viewing. The demo on its site explains that it can semi-automatically learn semantic relationships between pictures that may look different but that have some semantic similarities to suggest them as the ones viewers might want to see next, based on observed preferences (it compares this to the recommendations Amazon delivers about books based on readers’ past habits).

On the medical imaging front, THESEUS:Medico is a research program initiated by Germany’s Federal Ministry of Economy and Technology (BMWi) that says it aims at building up an intelligent, scalable and robust search engine for medical images. The goal is to provide semantic access to medical image databases, autonomously extracting knowledge from these databases to enable complex search queries, such as finding cases that have a similar appearance to a particular patient medical image. Partners in that consortium include Siemens AG, Ludwig Maximilans University, Erlangen University Hospital, the Fraunhofer Institute for Computer Graphics and the German Center for Artificial Intelligence. Siemens experts, the company says, are focusing initially on 3D data sets from tomography devices (CT/MR) in order to close the existing semantic gap in a predefined area between unstructured image data and medical terminology (where “semantic” refers to the ability of a computer program to understand image content).

But back to the Intel competition and Lynbrook High School. Maybe there’s something in the water there: One of the 300 semi-finalists, chosen from among 1,736 entrants representing 472 high schools in 44 states, the District of Columbia, Puerto Rico and four overseas schools, was Lynbrook’s 17-year-old Tony Ho, whose research project was entitled, “Ontology Driven Semantic Annotation Method for Public Microarray Repositories.”

The finalists will be gathering in D.C. in March to have a chance to display their work to the public and compete for an award of $100,000. We’ll be pulling for ya, David.