Computer Program Recognizes Sketches More Accurately Than a Human

Posted on July 21, 2015

Bird sketches test for Sketch-A-Net program

Scientists say they have developed a computer program that can recognize sketches with greater accuracy than a human. The program, called Sketch-a-Net, can recognize sketches 74.9% of the time. This is slightly better than humans who managed a sketch identification success rate of 73.1%.

The scientists from Queen Mary, University of London (QMUL), says the program identified the above images of a seagull, pigeon, flying bird and standing bird better than humans. The program's accuracy rate on these sketches was 42.5% compared to just 24.8% for humans. The research was presented at the British Machine Vision Conference.

Sketches have been used a communication tool for thousands of years. The abstract and varied nature of sketches and the fact that consist of black and white lines makes it a challenge for them to be identified with a computer program. The Sketch-a-Net program is described as a "deep neural network." It was designed to emulate the processing of the human brain. The researchers note the program could be useful in situation like matching an artist's sketch of a criminal to a mugshot in a police database.

Timothy Hospedales, co-author of the study and lecturer at the School of Electronic Engineering and Computer Science, QMUL, says in a statement, "It's exciting that our computer program can solve the task even better than humans can. Sketches are an interesting area to study because they have been used since pre-historic times for communication and now, with the increase in use of touchscreens, they are becoming a much more common communication tool again. This could really have a huge impact for areas such as police forensics, touchscreen use and image retrieval, and ultimately will help us get to the bottom of visual understanding."

The research paper can be found here (PDF file).

Photo: QMUL, Mathias Eitz, James Hays and Marc Alexa