engleski

Make Potatoes Great Again

Image matching by using keypoint detectors and descriptors is used in many image processing and computer vision tasks. While many detectors and descriptors result in accurate matching, they have difficulties on images that are blurry and out of focus like the ones obtained on the so called potato cameras. Most detectors and descriptors work on grayscale images and in this paper new grayscale channels are learned for maximizing the number of correct feature matches. In contrast to the usual grayscale channel, with the new channel the keypoints and descriptors produce up to 3% more correct matches on regular images and well over 100% more of them on low quality images without adding any additional computation cost. It is also shown how only using new channel grayscale definitions can improve the accuracy of ORB-SLAM2 without changing any of its other steps. The experimental results are presented and discussed.

blur ; grayscale ; keypoint descriptor ; keypoint detector ; image ; matching ; ORB-SLAM2 ; SLAM

nije evidentirano