Image Registration using Median Absolute Deviation –based Adaptive RANSAC

Résumé

Image registration encompasses topics such as change detection and remote sensing. Feature-based registration is one of the main approaches and relies on feature extraction and matching. The Oriented FAST and Rotated BRIEF (ORB) algorithm is one of the most robust methods used in feature registration. Random sample consensus (RANSAC) is an optimization method for reducing mismatches in ORB algorithm. However, RANSAC-based methods have certain deficiencies, including rapid increase in computational time, higher false positive ratio, and the need for an empirically determined fixed threshold value. The aforementioned shortcomings result in a reduction in the accuracy of the transform model parameters. In this paper, a modified RANSAC algorithm is proposed, incorporating a Median Absolute Deviation (MAD)-based adaptive threshold, to enhance the efficacy of the method. The threshold value is determined by the MAD of the distances between each point and its model-transformed counterpart. This method enhances the RANSAC algorithm, by taking into consideration the early best matches of each iteration, increasing the number of inliers, and looping through an iterative process based on least squares estimation. The simulation results show that the proposed method is robust to distortion and noise. The results demonstrate that the proposed approach outperforms standard ORB in terms of Mean Squared Error (MSE), Normalized Mutual Information (NMI), Structural Similarity Index Method (SSIM), and Peak Signal-to-Noise Ratio (PSNR).

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Voir l'article sur le site de la revue

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0