Data and Model-Driven Selection Using Color Regions

• Authors: Tanveer Fathima Syeda-Mahmood
• Venue: International Journal of Computer Vision
• Year: 1997

A key problem in model-based object recognition is selection,namely, the problem of determining which regions in the imageare likely to come from a single object. In this paper wepresent an approach that uses color as a cue to performselection either based solely on image-data (data-driven), orbased on the knowledge of the color description of the model(model-driven). Specifically, the paper presents a method ofcolor specification in terms of perceptual color categories andshows its relevance for the task of selection. The colorcategories are used to develop a fast region segmentationalgorithm that extracts perceptual color regions in images. Thecolor regions extracted form the basis for performing data andmodel-driven selection. Data-driven selection is achieved byselecting salient color regions as judged by a color-saliencymeasure that emphasizes attributes that are also important inhuman color perception. The approach to model-driven selection,on the other hand, exploits the color and other regioninformation in the 3d model object to locate instances of theobject in a given image. The approach presented tolerates someof the problems of occlusion, pose and illumination changes thatmake a model instance in an image appear different from itsoriginal description. Finally, the utility of color-basedselection is demonstrated by showing the extent of searchreduction possible when color-based selection is integrated witha recognition system.