This paper presents a new learning-based approach to estimate local homography of a given 3D object and shows that it ismore accurate than specific affine region detection methods. Unlike the previous works that attempt this by adapting iterative algorithms, our method introduces a direct estimation. It performs in three step. First, a training set of features captures geometry and appearance information about keypoints taken from multiple views of the object. Then incoming keypoints are matched against the database in order to retrieve a cluster of features representing their identity. Finally the retrieved clusters are used to estimate the local pose of the patches.
展开▼