Harris corner detector

Using that significant change exists to all direction in Derivative Filter at the corner

• Harris corner response is invariant to rotation (
  Eigenvalue is same)

• Harris corner detector is not invariant to scale

Error Function

for each point

1. pick small region
   Window Function or
   Gaussian Filter

2. compute derivatives

3. compute products of derivatives at every pixel

4. compute the sums of products of derivatives at each pixel and subtract it
   DC bias

5. Compute the covariance matrix (M 모양)

6. Compute
   Eigenvector and eigenvalues

7. Use threshold on a function of eigenvalues to detect corners

Pros

• Corner response R is invariant to image rotation

• R is Partial invariance to affine intensity change (shift and intensity scale)

Limitation

• 색변화만 감지한다

• not invariant to scale so we need multi-scale detection (can’t auto select)

• 주변만 고려한다