Read the associated paper on Methods to improve SIFT
Scale invariant feature transform for image processing. This is an investigation of methods to improve SIFT algorithms.
- Images are first transformed into a collection of vectors.
- Keypoints (of high contrast) for distinguishing features in the image are identified and stored in a DB
- Some unstable points can falsely record as keypoints. We avoid this using keypoint localization
- The resulting feature vector is denoised. It's much smaller and more accurate.
Keypoint localization eliminated a significant amount of noise. Varying the image's contrast produced more accurate results, even with keypoint localization.
Shruti Appiah, Mira Sleiman
Copyright © 2017 Shruti Appiah