Combining rejection-based pattern classifiers
L. Mascarilla, Carl Frélicot · 2002
This article deals with the combination of the first stage of two-fold rejection-based classifiers for pattern classification. This Dempster-Shafer's model-based combination uses some relevant characteristics of the different two-stage classifier strategies we have identified. These strategies differ on the managing of the ambiguity and distance rejection (independently or not). We propose some clever basic probability assignments to reject classes before using the combination rule. After combination, a decision rule is proposed for classifying or rejecting patterns. We emphasize that such rejection is not related to a lack of consensus between the classifiers but to the initial reject options. In case of ambiguity rejection, a class-selective approach has been used. Some illustrative results on artificial data are given.