IGFTT: towards an efficient alternative to SIFT and SURF

The invariant feature detectors are essential components in many computer vision applications, such as tracking, simultaneous localization and mapping (SLAM), image search, machine vision, object recognition, 3D reconstruction from multiple images, augmented reality, stereo vision, and others. However, it is very challenging to detect high quality features while maintaining a low computational cost. Scale-Invariant Feature Transform (SIFT) and Speeded-Up Robust Features (SURF) algorithms exhibit great performance under a variety of image transformations, however these methods rely on costly keypoint’s detection. Recently, fast and efficient variants such as Binary Robust Invariant Scalable Keypoints (BRISK) and Oriented Fast and Rotated BRIEF (ORB) were developed to offset the computational burden of these traditional detectors. In this paper, we propose to improve the Good Features to Track (GFTT) detector, coined IGFTT. It approximates or even outperforms the state-of-art detectors with respect to repeatability, distinctiveness, and robustness, yet can be computed much faster than Maximally Stable Extremal Regions (MSER), SIFT, BRISK, KAZE, Accelerated KAZE (AKAZE) and SURF. This is achieved by using the search of maximal-minimum eigenvalue in the image on scale-space and a new orientation extraction method based on eigenvectors. A comprehensive evaluation on standard datasets shows that IGFTT achieves quite a high performance with a computation time comparable to state-of-the-art real-time features. The proposed method shows exceptionally good performance compared to SURF, ORB, GFTT, MSER, Star, SIFT, KAZE, AKAZE and BRISK.