Efficient homography estimation method for panorama

In this paper, we propose a fast and efficient method for estimating a homography. The homography is used to generate a stitched image such as UHD (Ultra-High-Definition) image from multiple HD (High-Definition) images. The homography estimation is the most important part in stitching techniques because the accuracy of the estimated homography means confidence of correspondence between the HD images. Specifically, we perform the feature point extraction and the outlier removal to estimate a homography. However, when too many features exist in the images, we require lots of computation, to estimate a homography and the outlier removal. Furthermore, there is a high probability of obtaining the poor homography. Therefore, we proposed a method to extract the correspondences between down-sampled images and estimate the homography H´ and then compensate it to estimate the original one. In order to evaluate the proposed algorithm, comparison results are shown with the traditional stitching methods. It is believed that the proposed algorithm can be a useful tool to the image stitching fields.