Distinctive Image Features from Scale-Invariant Keypoints
Citations
94 citations
Cites background from "Distinctive Image Features from Sca..."
...The arrow illustrates the increasing space scales [15]....
[...]
...Following the suggestion in [15], k is set to √ 2, which leads to a significant difference in successive scales....
[...]
...Scale-space extrema in the Difference of Gaussians are regarded as the most stable scale-invariant features [15]....
[...]
...the scale-invariant feature transform (SIFT) [11], [13]–[15]....
[...]
94 citations
Cites methods from "Distinctive Image Features from Sca..."
...Given their scale and local affine invariance properties, we opt to use SIFT [15] or SURF [16] instead, as they constitute a better option for matching visual features...
[...]
...SIFT and SURF descriptor matching are quite reliable in many situations, yet RANSAC is needed to eliminate outliers due to erroneous stereo and temporal matching, as outliers are capable of introducing large error into the solution....
[...]
...Given their scale and local affine invariance properties, we opt to use SIFT [15] or SURF [16] instead, as they constitute a better option for matching visual features from varying poses....
[...]
...To deal with scale and affine distortions in SIFT, for example, keypoint patches are selected from difference-of-Gaussian images at various scales, for which the dominant gradient orientation and scale are stored....
[...]
...Our technique produces similar results whether we use SIFT or SURF, with SURF running significantly faster....
[...]
94 citations
Cites background or methods from "Distinctive Image Features from Sca..."
...The reasons for new detections are explained in [3]....
[...]
...2) Measuring Matching Performance: Two keypoints are considered to be a match iff the Euclidean distance between their SIFT descriptors is below a certain threshold λ [3], [7]....
[...]
...The original SIFT algorithm that is proposed by Lowe [3], despite being invariant to rotations on the plane P2 , is unable to handle the projective transformations due to camera rotation [14]....
[...]
...The scale-invariant feature transform (SIFT) [3] is arguably one of the most popular matching algorithms, being broadly used in robotics because of its invariance to common image transformations such as scale, rotation, and moderate viewpoint change [4], [5]....
[...]
93 citations
Cites background or methods from "Distinctive Image Features from Sca..."
...However, even in the original work on SIFT descriptor matching (Lowe 2004) it is shown that the similarity of the descriptors is not only dependent on the distance of the descriptors, but also on the location of the features in the feature space....
[...]
...Most, if not all, recent stateof-the-art methods extend the bag-of-words representation introduced by Sivic and Zisserman (Sivic and Zisserman 2003) who represented the image by a histogram of “visual words”, i.e., discretized SIFT descriptors (Lowe 2004)....
[...]
...To avoid bias (by quantization errors, for example), instead of using the vector-quantized form of the descriptors, the conventional image matching (based on the full SIFT (Lowe 2004)) has to be used....
[...]
...4.2 Feature Tracks To avoid bias (by quantization errors, for example), instead of using the vector-quantized form of the descriptors, the conventional image matching (based on the full SIFT (Lowe 2004)) has to be used....
[...]
93 citations
References
46,906 citations
16,989 citations
13,993 citations
7,057 citations
3,422 citations