Pete Ringwood home

Error Level Analysis

2010 - 2012

Filed under: Image processing, fraud detection

Note: After two years, the error level analysis servers have been turned off, to be moved on to other projects.

Error Level Analysis was a project run from 2010 to 2012 to algorithmically determine the authenticity of digital images. It would take an image, process it, and return a copy of with the error levels visualised throughout.

In the following example image, the process shows differing error levels throughout, evident by the differing areas of brightness throughout the resulting image. This strongly suggests some form of digital manipulation. Areas to note are the lips and shirt of the models, as well as the eyes. All are at significantly different error levels than their surroundings - they have been digitally manipulated. An image which has had no manipulation should essentially be a single colour - all parts of the image should at a single error level.

Original image from valleygirl.com.au

The website was freely available as a service from 2010 to 2012, at which stage it was closed down to move on to other projects. Over that period, 290,000 images were processed with the results served to over 1.5 million people.