In order to create a seamless and seemingly natural panorama, we propose a novel stitching method when a panoramic scene contain two predominate planes. Firstly, compute each homography of per planes. Then, how to set the each of weight in dual-homography become an important step. The traditional method of setting weights is to directly calculate European distance between original image location pixel points and feature points. The disadvantage is the weights of singular points seriously impact the overall decision. In this paper, we proposed a static probability model of error matching to optimize weights by multivariate student’s t distribution. No only error matching probability, but also error amount and distance of feature points are all considered in the weight model. Finally, a renewal single homography is defined by establishing contact between dual-homography and weights. Experiments show the homography matrix is more robust and accurate to perform a nonlinear warping. The proposed method is easily generalized to multiple images, and allows one to automatically obtain the best perspective in the panorama.
|