Computer Program Identifies Rough Sketches
CREDIT: Brown University
A new computer program can identify doodles of objects almost as well as humans.
The advance could help pave the way for sketch-based interfaces and search applications, say the software's developers from Brown University and the Technical University in Berlin.
The researchers taught their computer program to learn how to recognize doodles by feeding 20,000 sketches of 250 object categories that were hand-drawn by humans. They then gave the program new sketches, and asked it to identify them in real time, as a person was drawing them.
The program was able to successfully identify sketcheswith around 56-percent accuracy, as long as the object being sketched was part of one of the 250 categories it had seen previously. For comparison, humans managed to correctly identify the sketches about 70 percent of the time.
“The gap between human and computational performance is not so big, not as big certainly as it is in other computer vision problems,” team member James Hays, a computer scientist at Brown, said in a statement.
The team hopes to expand the software's capabilities by turning it into a gaming app that can collect the data that human players input.
"The algorithm really needs to see close to 100 instances of how people draw lions, and then it becomes possible to tell lions from potted plants,” Hays said.