An Improved RANSAC Method Using Absolute Median Deviation (MAD) Based on Adaptive Threshold
Khalid Akdim, Ahmed Roukhe, Hassane Roukhe · 2024
Image processing now plays a relevant role in a wide range of fields. Image registration, a critical aspect of image processing, encompasses essential issues like image fusion, and remote sensing. One of the most robust algorithms used in feature registration is the Oriented FAST and Rotated BRIEF (ORB). Several optimization methods, including random sample consensus (RANSAC), have been adopted to reduce the outliers in this algorithm ORB and find optimal parameters of the image registration problem. However, RANSAC-based methods exhibit some deficiencies such as a high number of iterations, increased false positives ratio of mismatches, and a fixed threshold value determined through experimentation resulting in lower accuracy of transform model parameters. In this paper, we propose a modified version of RANSAC incorporating a Median Absolute Deviation (MAD) based on an adaptive threshold to address these deficiencies. To verify the validity of our proposed approach, a comprehensive experimental analysis has been done on a database of two images. A study, of the influence of certain factors such as the angle of rotation of the sensed image, the scale of the sensed image, and the noise applied to the sensed image. The simulation results show that the proposed method can achieve good transform model parameters in terms of Mean Square Error (MSE), Structure Similarity Index Method (SSIM), and Peak Signal to Noise Ratio (PSNR).