Boosted Image Classification: An Empirical Study
Nicholas R. Howe · Smith ScholarWorks (Smith College) · 2002
The rapid pace of research in the fields of machine learning and image comparison has produced powerful new techniques in both areas. At the same time, research has been sparse on applying the best ideas from both fields to image classification and other forms of pattern recognition. This paper combines boosting with stateof-the-art methods in image comparison to carry out a comparative evaluation of several top algorithms. The results suggest that a new method for applying boosting may be most effective on data with many dimensions. Effectively marrying the best ideas from the two fields takes effort, but the techniques and analyses developed herein make the task straightforward.