Using colour information for object recognition
Teaching computers to perform ‘human’ tasks has become a common theme in an era of increased automation and a quest to reduce potential human errors in a variety of processes. Automated colour recognition has important use in fields as diverse as robotics, gaming and photo editing. In order to develop new methods combining colour and shape information into algorithms for object recognition, EU-funded researchers initiated the ‘Task specific description of visual color information’ (TS-VICI) project. Scientists sought to develop an automatic adaptation colour description combining photometric invariants (features repeatable with respect to changes in lighting) with discriminative power (distinctness, or the usefulness of a feature for classification purposes). In addition, researchers suggested adaptations to the colour descriptor incorporating task-specific or class-specific information for higher performance – for example, colour descriptors help identify colour-invariant objects such as flamingos but do not add value to recognition of colour-variant objects such as cars. Overall, results suggested that the use of colour to guide shape description outperformed existing methods using a combination of colour and shape information. Off-shoots of the project outcomes included research on physically realistic recolouring of objects such as required by photo editing, publicity and gaming applications. The TS-VICI project made valuable enhancements to theories and algorithms associated with automated image analysis and object recognition valuable to robotics, computer vision and photo editing with important potential for commercial exploitation.
