Models that captures the common structure of anobject class have appeared few years ago in the literature(Jojic and Caspi in Proceedings of IEEE Computer SocietyConference on Computer Vision and Pattern Recognition(CVPR), pp. 212–219, 2004; Winn and Jojic in Proceedingsof International Conference on Computer Vision (ICCV),pp. 756–763, 2005); they are often referred as “stel models.”Their main characteristic is to segment objects in clear, oftensemantic, parts as a consequence of the modeling constraintwhich forces the regions belonging to a single segment tohave a tight distribution over local measurements, such ascolor or texture. This self-similarity within a region in a singleimage is typical of many meaningful image parts, evenwhen across different images of similar objects, the correspondingparts may not have similar local measurements.Moreover, the segmentation itself is expected to be consistentwithin a class, although still flexible. These models havebeen applied mostly to segmentation scenarios.In this paper, we extent those ideas (1) proposing to capturecorrelations that exist in structural elements of an imageclass due to global effects, (2) exploiting the segmentationsto capture feature co-occurrences and (3) allowing the useof multiple, eventually sparse, observation of different nature.In this way we obtain richer models more suitable torecognition tasks.
Stel Component Analysis: joint segmentation, modeling and recognition of objects classes
Alessandro Perina;Marco Cristani;Vittorio Murino
2012-01-01
Abstract
Models that captures the common structure of anobject class have appeared few years ago in the literature(Jojic and Caspi in Proceedings of IEEE Computer SocietyConference on Computer Vision and Pattern Recognition(CVPR), pp. 212–219, 2004; Winn and Jojic in Proceedingsof International Conference on Computer Vision (ICCV),pp. 756–763, 2005); they are often referred as “stel models.”Their main characteristic is to segment objects in clear, oftensemantic, parts as a consequence of the modeling constraintwhich forces the regions belonging to a single segment tohave a tight distribution over local measurements, such ascolor or texture. This self-similarity within a region in a singleimage is typical of many meaningful image parts, evenwhen across different images of similar objects, the correspondingparts may not have similar local measurements.Moreover, the segmentation itself is expected to be consistentwithin a class, although still flexible. These models havebeen applied mostly to segmentation scenarios.In this paper, we extent those ideas (1) proposing to capturecorrelations that exist in structural elements of an imageclass due to global effects, (2) exploiting the segmentationsto capture feature co-occurrences and (3) allowing the useof multiple, eventually sparse, observation of different nature.In this way we obtain richer models more suitable torecognition tasks.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.