This paper aims to explore frequency behavior of isotropic (regular SIFT) and anisotropic (Bi-SIFT and Tri-SIFT) versions of the scale-space keypoint detection algorithm SIFT. We introduced a new smoothing function Trilateral filter that can be used in formation of a scale-space as an alternative to the Gaussian scale-space. The number of matching pixels, warping error, and scatteredness are employed in comparison. We made the comparison out of face dataset and object dataset for scale, orientation, and view-angle transformations as well as lighting and compression variations. The comparison results show that anisotropic smoothing detects more keypoints than isotropic one. The Tri-SIFT is more robust to variation in viewpoint angle.

+More

scale orientation and viewangle transformations lighting and compression gaussian classemphasistypeitalic trilateral scalespace keypoint detection algorithm object dataset matching pixels isotropic viewpoint keypoints frequency behavior trisift isotropic regular sift and anisotropic bisift and trisift versions anisotropic smoothing

Ömer Muhammet Soysal,Kazim Şekeroğlu,.Comparison of SIFT, Bi-SIFT, and Tri-SIFT and their frequency spectrum analysis. 28 (8),875–902.

Ömer Muhammet Soysal,Kazim Şekeroğlu,"Comparison of SIFT, Bi-SIFT, and Tri-SIFT and their frequency spectrum analysis" 28

Ömer Muhammet Soysal,Kazim Şekeroğlu,"Comparison of SIFT, Bi-SIFT, and Tri-SIFT and their frequency spectrum analysis"