Reliable interest point detection under large illumination variations

This appendix proves the illumination robustness of the proposed corner detector over the Harris corner detector. I. HARRIS CORNER DETECTOR The Harris corner detector [1] is based on the autocorrelation matrix of the image gradients. The autocorrelation matrix A(x, y; I) of an image I at a pixel (x, y) is given as follows: A(x, y; I) =     (m,n)∈N ∂ ∂x I (m, n) 2 (m,n)∈N ∂ ∂x I (m, n) ∂ ∂y I (m, n) (m,n)∈N ∂ ∂x I (m, n) ∂ ∂y I (m, n) (m,n)∈N ∂ ∂y I (m, n) 2     , (1) where ∂ ∂x and ∂ ∂y calculates the gradients in horizontal and vertical directions, respectively; and N is a set of pixels around (x, y). Usually a weighted sum of the gradients in N is taken using a Gaussian function centered at (x, y) to give more weight to the pixels that are close to (x, y). The cornerness response function of the Harris corner detector is based on the determinant and trace of the autocorrelation matrix: Rharris(x, y) = detA(x, y; I) − ...

• Type: Conference
• Year: 2008
• Where: ICIP
• Authors: Murat Gevrekci, Bahadir K. Gunturk
• Keywords: Autocorrelation Matrix, Contrast Invariant Feature, Harris Corner Detector, ICIP 2008, Image Processing