Illumination-invariant color object recognition and novel color segmentation techniques.

Abstract

With the fast development of computer technology, computer vision has been widely exploited to recognize objects in various environments without human intervention. The tasks of object recognition system include identifying the ob-jects in the scene and determining the region occupied by these objects. Gen-erally, there are two factors that will influence the performance of the color object recognition: 1) the variation in environmental illuminations; 2) the ac-curacy of the extracted boundary of interesting regions and objects. Based on these two factors, in this thesis, research has been devoted on improving the performance of color object recognition by extracting illumination-invariant features and proposing novel color segmentation techniques. These novel color segmentation techniques involved two aspects: 1) from perceptually uniform color distance—measuring the color distance between each pixel and other ref-erence points, 2) from object material— computing the reflectance spectrum of object's surface.