کد مقاله | کد نشریه | سال انتشار | مقاله انگلیسی | نسخه تمام متن |
---|---|---|---|---|
406357 | 678081 | 2015 | 13 صفحه PDF | دانلود رایگان |
Usual multi-class classification techniques often rely on the availability of all relevant features. In practice, however, this requirement restricts the type of features that can be considered. Features whose value depends on some partial, intermediate classification results, can convey precious information but their nature hinders their use. A typical example is the identification of objects in a scene, where the distance from some yet unclassified object to some other that would already be identified earlier in the process. This paper proposes a generic method that solves classification problems involving such features in an incremental way. It proceeds by decomposing the multi-class problem into a sequence of simpler binary problems. Once a binary classifier gives an object its class tag, all features depending on this object are computed and appended to the list of known features. Experiments with both synthetic and real data, comprised of tomographic images, show that the proposed method is effective.
Journal: Neurocomputing - Volume 152, 25 March 2015, Pages 45–57