Object Recognition Using Hierarchical SVMs

The paper deals with the object recognition problem. The objective is to localize and recognize the known objects in different orientations on cluttered background. As a learning tool we choose support vector machines (SVMs). To eliminate the problem caused by cluttered background we organize the image pixels in tree structures, which enable us to deal only with the object pixels. Both, oneand two-class SVMs are combined in the recognition process. One-class SVMs, used at the first stage, allow us to avoid the “nonobject” class generation as required by two-class SVM for object localization task. Two-class SVMs are applied to further resolve the recognition process when necessary. As demonstrated by experimental results the proposed method reduces the number of erroneously recognized objects.