Contour features play an important role in object recognition. Psychological experiments have shown that maximum-curvature points are most distinctive along a contour. This paper presents an object detection method based on Principal Contour Fragments (PCFs), where PCFs are extracted by partitioning connected edge pixels at maximum-curvature points. An object is represented by a set of PCFs and their mutual geometric relations. The mutual geometric relations are described in each PCF's local coordinate system, and they are invariant to translation, rotation, and scale. With this representation, given any individual PCF, the system is capable of predicting all other PCFs' geometric properties. Object instances are detected in test images by sequentially locating PCFs whose geometric properties best match their predictions. Detected objects are verified according to their similarity to the model based on both individual PCF descriptors and mutual relation descriptors. Evaluation results show that the system works well in the presence of background clutter, large scale changes, and intra-class shape variations.

PDF

In Canadian Conference on Computer and Robot Vision (CRV-11) 2011.

@inproceedings{xu:crv11b, title={Object Detection Using Principal Contour Fragments}, author={Changhai Xu and Benjamin Kuipers}, booktitle={Canadian Conference on Computer and Robot Vision (CRV-11)}, url="http://www.cs.utexas.edu/users/ai-lab?xu:crv11b", year={2011} }