Peter Auer | Search PubMed

PubMed2006

Generic object recognition with boosting.

This paper explores the power and the limitations of weakly supervised categorization. We present a complete framework that starts with the extraction of various local regions of either discontinuity or homogeneity. A variety of local descriptors can be applied to form a set of feature vectors for each local region. Boosting is used to learn a subset of such feature vectors (weak hypotheses) and to combine them into one final hypothesis for each visual category. This combination of individual extractors and descriptors leads to recognition rates that are superior to other approaches which use only one specific extractor/descriptor setting. To explore the limitation of our system, we had to set up new, highly complex image databases that show the objects of interest at varying scales and poses, in cluttered background, and under considerable occlusion. We obtain classification results up to 81 percent ROC-equal error rate on the most complex of our databases. Our approach outperforms all comparable solutions on common databases.

Algorithms↗

PubMed2002

Identification of regional varieties by intonational cues. An experimental study on Hamburg and Berlin German.

Two experiments examined the commonly held belief that regional varieties of German can be identified by intonational features alone. In both experiments, listeners were presented with regional intonational contours of German. In the first experiment, listeners judged contours of Hamburg urban vernacular compared with contours of Northern Standard German. In the second experiment, listeners judged contours of Berlin urban vernacular compared with contours of both Northern Standard German and Low Alemannic German. The performance of listeners was found to vary with their linguistic experience. Listeners who were familiar both with the local variety and with some nonlocal variety by personal contact performed better than listeners who were familiar with the local variety only. Moreover, also listeners not familiar with Hamburg German and Berlin German, respectively, were found to perform the identification test with some success. This led to the conclusion that overall success rates do not only depend on true recognition of local contours but may additionally be enhanced by using some kind of elimination strategy. A second factor that affected performance was the choice of speaker for generating the carrier utterances. In the first experiment, all carrier utterances were produced by a speaker of Northern Standard German. In the second experiment, two sets of carrier utterances were used. The first set was obtained from a speaker of Northern Standard German and the second set from a speaker of Berlin urban vernacular. As expected, Berlin contours were better identified when presented with an utterance that was produced by a speaker of Berlin urban vernacular. However, no uniform effect was found for the different contours that were examined.

Berlin↗