SIFT Descriptor

Idea of descriptor

Represent the image content as a "constellation" of local features that are invariant to translate, scale, rotation , and other image parameters.

Overall SIFT Procedure

1. Scale-space extrema detection

2. Key Point Localization

3. Orientation Assignment

The base orientation is the dominant direction of the gradient.

1. To localize orientation to a feature we create a histogram of local gradient directions and their resulting magnitudes at a selected scale - 36 bins.

2. The canonical orientation is assigned to the peak of the smoothed histogram.

3. Each feature points has some properties: its $(x,y)$ coordinates, and an invariant scale and orientation.

4. Keypoint Description

A descriptor is distinctive and invariant. Use local image gradients at selected scale and rotation to describe each keypoint region.

1. Normalize:

   Rotate a keypoint's window based on the standard orientation.

   Then scale the window size based on the keypoint's scale.

2. Create a feature vector based upon:

   • A histogram of gradients, which we determined previous when finding the orientation
   • Weighted by a centered Gaussian Filter, to appropriately value the center gradients more.

3. Reduce effect of illumination

   • Clip gradient magnitudes to avoid excessive influence of high gradients.