---------------------------------------------------------------------- ARTIFICIAL INTELLIGENCE RESEARCH SEMINAR Department of Computer Science Worcester Polytechnic Institute SPEAKER: Prof. Mark Stevens TITLE: Color classification: using object color for segmentation DATE: Thursday, TIME: 11 am ROOM: FL246, Beckett Conference Room. ABSTRACT: Color is an important component influencing the appearance of an object in an image. This talk will address techniques (and possible avenues of collaborative research) for using color as a cue for segmentation. Our past work has used decision trees to created linear planes partitioning RGB color space into regions representing the color of an object under a variety of illumation conditions. The decision tree (DT) mechanism has performed well on several simple problems involving colored cubes in indoor settings as well as complex outdoor scenes involving highly camouflaged military vehicles. Comparisons to a neural network (NN) classifier have been difficult to interpret (no clear winner). The question becomes is this an easy pattern recognition problem? or are the two techniques (DT and NN) learning similar partitions? We believe that it is a fairly difficult problem to compare classifiers due to the large number of parameters and tuning that go into making either algorithm work well. Finally how do the techniques of boosting and bagging affect such a comparison? ---------------------------------------------------------------------- Everyone is welcome to attend!