Image analysis test pits man versus machine

Working with colleagues from Johns Hopkins University, François Fleuret, a senior scientist at EPFL (Lausanne, Switzerland) has developed a test to quantify how inferior computer-based image-processing systems are to humans when it comes to grouping patterns in digital images.

Nov 8th, 2011

Working with colleagues from Johns Hopkins University, François Fleuret, a senior scientist at EPFL (Lausanne, Switzerland) has developed a test to quantify how inferior computer-based image-processing systems are to humans when it comes to grouping patterns in digital images.

In the test, the researchers compared the efficiency of human and machine learning systems in assigning an image to one of two categories determined by the spatial arrangement of its constituent parts.

The experiments demonstrated that while human subjects grasped the principles of separating the images from a handful of examples, the error rates of computer programs fluctuated wildly and remained far behind that of humans even after exposure to thousands of examples.

The EPFL researchers say that this is the first time that the performance of classical learning algorithms and humans has been quantified in such a way.

More information on the research can be found in the Proceedings of the National Academy of Sciences.

-- By Dave Wilson, Senior Editor, Vision Systems Design

More in Home