Using Segmentation to Verify Object Hypotheses
We present an approach for object recognition that combines detection and segmentation within a efficient hypothesize/test framework. The current state-of-the-art (for finding, say, faces, cars, or pedestrians) seems to be window-based classifiers. Such approaches can presumably be hindered by their lack of explicit encoding of object shape/structure -- one might, for example, find faces in trees. We adopt the following strategy; we first use these systems to generate possible object locations (by tuning them for low missed-detections and high false-positives). At each hypothesized detection, we compute a local segmentation (using a window of slightly larger extent than that used by the classifier). We learn from training data those segmentations that are consistent with true positives. We then prune away those hypotheses with bad segmentations. We show this strategy often leads to significant improvements over established approaches (such as ViolaJones and leading competitors on the PASCAL challenge).
Text ReferenceDeva Ramanan. Using segmentation to verify object hypotheses. In CVPR, 1–8. 2007.
AUTHOR = "Ramanan, Deva",
TAG = "object_recognition,people",
TITLE = "Using Segmentation to Verify Object Hypotheses",
BOOKTITLE = "CVPR",
YEAR = "2007",
PAGES = "1-8",
BIBSOURCE = "http://www.visionbib.com/bibliography/applicat814.html#TT61549"