Supervised Learning Based Stereo Matching Using Neural Tree

摘要

In this paper, a supervised learning based approach is presented to classify tentative matches as inliers or outliers obtained from a pair of stereo images. A balanced neural tree (BNT) is adopted to perform the classification task. A set of tentative matches is obtained using speedup robust feature (SURF) matching and then feature vectors are extracted for all matches to classify them either as inliers or outliers. The BNT is trained using a set of tentative matches having ground-truth information, and then it is used for classifying other sets of tentative matches obtained from the different pairs of images. Several experiments have been performed to evaluate the performance of the proposed method.