To search, Click below search items.


All Published Papers Search Service


Image Augmentation for Keypoint Detection and Matching Assessments


Ibrahim El rube


Vol. 22  No. 11  pp. 557-566


Keypoint detection and matching algorithms are frequently compared in the literature using datasets of real-world images that have a range of geometric and non-geometric variations; these include viewpoints, illuminations, visual content, and distortions. Homography (H) matrices often describe geometric variations when utilizing these image datasets. However, models for non-geometric differences between these images are rarely offered, resulting in inaccurate and misleading comparisons. This study presents a methodology for objectively comparing classical keypoint detection and matching algorithms by eliminating implicit non-geometric influences from assessments, therefore, offering a step towards limiting the comparison between an image pair to the geometric transformations between them. This proposed technique uses the H matrix provided by the image dataset to generate an augmented image that resembles one of the images in each image group. The performance of the proposed technique was evaluated using several traditional keypoint detections and matching techniques using image groups from well-known datasets to determine the impact of excluding non-geometric changes. The assessments are conducted using the performance measures of repeatability, precision, and recall rates.


Augmented image; geometric transformation; homography matrix; keypoints.