July 24, 2015 at 11:48For the first time, a program better than a person copes with pattern recognition
The Sketch-a-Net program developed at Queen Mary’s University of London was able to correctly interpret the picture with the image of a bird in 74.9% of cases, having coped with this task better than the experimental people (73.1%).
The work, announced at the 26th British Machine Vision Conference (September 2015, Swansea), also showed that Sketch-a-Net was better than a person in managing detailed pattern recognition. For example, she successfully distinguished between a seagull, a flying bird, a standing bird, and a pigeon in 42.5% of cases, while people achieved an accuracy of only 24.8%.
Sketch-a-Net is based on a “deep neural network” that mimics the functioning of the human brain. When recognizing it, it even takes into account the drawing order of the lines that make up the drawing.
This technology can find application in a more rapid development of sign language when working with a computer. Or a more accurate image search.
The work, announced at the 26th British Machine Vision Conference (September 2015, Swansea), also showed that Sketch-a-Net was better than a person in managing detailed pattern recognition. For example, she successfully distinguished between a seagull, a flying bird, a standing bird, and a pigeon in 42.5% of cases, while people achieved an accuracy of only 24.8%.
Sketch-a-Net is based on a “deep neural network” that mimics the functioning of the human brain. When recognizing it, it even takes into account the drawing order of the lines that make up the drawing.
This technology can find application in a more rapid development of sign language when working with a computer. Or a more accurate image search.