In time, 3D construction of the Eiffel Tower, catching criminals and discovering information about everyday objects and people may become possible with a single photograph.
These possibilities may come to fruition with the help of the Department of Computer Science, which received a three-year $488,576 grant from the Department of Defense to research a large-scale image retrieval system and its applications.
“This will make the world better and safer,” said Hongchi Shi, chair of the Computer Science Department. “A lot of people can benefit from this.”
Yijuan Lu, assistant professor of the Department of Computer Science, is heading the image retrieval research with Qi Tian, associate professor of the Department of Computer Science at the University of Texas at San Antonio. They will work together with students from both universities to perfect what is called “content-based image retrieval (CBIR).
Image retrieval applications such as Google Goggles have popped up on both the iPhone and Android markets as a convenient way to search for information on the Internet by sending in a photograph of an object.
The Google Goggles application recognizes textual features in images the user captures. Goggle Goggles then searches the web for similar objects and photographs and returns the information to the user. This type of search is known as a partial-duplicate image search.
Lu plans to find a way to perfect this type of application and take it to a higher level. Her research is focused on image retrieval searches including forensics, video surveillance, automatic 3D construction and an image encyclopedia.
Law enforcement will be able to use the program to help identify suspects with unique markings such as tattoos or scars. Police will be able to send the image into the program and receive information about a tattoo’s meaning, the suspect’s affiliation with illegal groups and their address.
Lu’s program will help with the prevention of online copyright crimes by searching the a database. The program will then inform users whether or not a work is copyrighted or patented.
The program is not just aimed at aiding law officials. It will include an “image encyclopedia” which will allow users to send images to the program and receive information about them.
“If you go to a museum or travel to famous places, just use your cell phone to take a picture and our system can go through the Internet and try to find similar images and get more information about the image,” Lu said.
The program will also help in the construction of 3D models. Normally pictures from every angle of an object would have to be taken and then meshed together to make a model. With Lu’s program the user will be able to search for the images of the object to form a 3D model.
With such a large amount of data on the Internet, there are many problems that can occur with large-scale searches. The search may turn up with wrong images or be slowed by an overload of information.
“The user hopes to return the images within one second, which means we cannot compare these images pair by pair,” Lu said. “We need to design a more efficient algorithm.”
Tian will meet this week with Microsoft representatives to discuss the objectives of his and Lu’s research.
“This work has a lot of potential in the industry,” Tian said.