Object Detection Based on Multi-Scale Contour Fragments
In this paper, the authors present a novel object detection scheme using the multi-scale contour fragments. The template fragments are extracted by decomposing the template contour. The multi-scale hinge angle, contour direction and Partial Hausdorff Distance (PHD) are used to select candidates in the edge image. Then, the matches with different scales and directions are selected by the Multiclass Discriminative Field (MDF) from the candidates. With the matches and their corresponding sample fragments, the contours of the objects can be obtained. The experiments on their postmark dataset and the ETHZ dataset show that the proposed scheme is robust to detect a class of objects with different scales, directions and complex background.