Deep correlation based homography estimation for image stitching

초록

Homography estimation is essential in many image transformation cases. Currently, there are two different paradigms, specifically non-learning-based and learningbased methods. However, both methods are still prone to erroneous homography estimation. To address this problem, we introduce a deep correlation based homography estimation that provides more stable and accurate results. The main novelty of this paper is to consider the correlation for each extracted feature, which contributes to removing redundant features and increasing the accuracy. Experimental results show the usability of the proposed method to produce more reliable stitched images. The proposed method achieves state-of-the-art results compared to the previous methods. © 2021 ICIC International. All rights reserved.

키워드