Comparative Analysis of Feature Fusion Techniques for Enhancing RGB-IR Image Matching

This paper investigates different feature fusion techniques for enhancing RGB-NIR image matching. The objective is to evaluate the accuracy and execution time of these techniques in order to identify the most effective approach. Specifically, the techniques evaluated include Different combinations between SURF, SIFT, AKAZE, BRISK and ORB. The evaluation was conducted using samples from RGB-NIR Scene Dataset and performance was measured based on the number of inliers, repeatability, precision, recall and execution time. The results show that AKAZE-BRISK achieved the highest number of inliers, followed closely by ORB-BRISK with a comparable execution time. ORB-AKAZE achieved the best execution time with a sufficient number of inliers. These findings suggest that these techniques are all viable options for enhancing RGB-NIR image matching, depending on specific requirements. Overall, this study provides valuable insights into the strengths and limitations of different feature fusion techniques for enhancing RGB-NIR image matching using the RGB-NIR Scene Dataset as a sample dataset.