A. Summary (mostly come from from ABSTRACT)
This paper presents a method for extracting distinctive invariant features from images that can be used to perform reliable matching between different views of an object or 'scene'.
major stages of computation used to generate the set of image features:
1. Scale-space extrema detection:
searches over all scales and image locations by using DoG
2. Keypoint localization
3. Orientation assignment (based on local image gradient directions)
4. Keypoint descriptor
The features are invariant to image scale and rotation, and are shown to provide robust matching across a a substantial range of affine distortion, change in 3D viewpoint, addition of noise, and change in illumination.
The features are highly distinctive, in the sense that a single feature can be correctly matched with high probability against a large database of features from many images.
This paper also describes an approach to using these features for object recognition.
The recognition proceeds by matching individual features to a database of features from known objects using a fast nearest-neighbor algorithm, followed by a Hough transform to identify clusters belonging to a single object, and finally performing verification through least-squares solution for consistent pose parameters.
This approach to recognition can robustly identify objects among clutter and occlusion while achieving near real-time performance.
B. Conclusion (rewrite from CONCLUSION, plus some comments)
1. Distinctiveness, so that enables the correct match for a keypoint to be selected from a large database of other keypoints.
2. Invariant to image rotation and scale.
Achieve scale-invariant by apply various Gaussian kernel, and rotation-invariant by dominant gradient.
3. Robust across a substantial range of affine distortion, addition of noise, and change in
illumination.
4. Large numbers of keypoints leads to robustness in extracting small objects among clutter.
Also, use bags of feature points comes with time issue when matching keypoints among large-scale database.
5. The research of Weber, Welling, and Perona (2000) and Fergus, Perona, and Zisserman (2003) has shown the potential of this approach by learning small sets of local features that are suited to recognizing generic classes of objects.
張貼留言