Enhancement of Region of Interest Using Superpixel-Based Feature Matching
Jaehyun Ko, Deokwoo Lee · Journal of Korea Multimedia Society · 2024
A novel approach integrating Superpixel Segmentation with feature matching is proposed to enhance object recognition and segmentation. Traditional segmentation techniques often depend on pre-trained models, which can struggle with unlabeled objects or require extensive training. To overcome these limitations, the method begins with Feature Matching using SuperGlue model to identify keypoints be- tween images. Next, the SLIC algorithm is applied to Superpixel Segmentation, dividing the image into smaller, coherent regions. The identified keypoints are then linked to their corresponding superpixels, allowing these regions to be grouped and expanded into significant Regions of Interest (ROI). This ap- proach facilitates not only effective image segmentation but also emphasizes key objects within the scene, enabling a more precise and adaptable segmentation process. The method proves effective in scenarios requiring accurate object recognition and segmentation, particularly when dealing with un- familiar or unlabeled objects.