We present in this section a generalization of a gray level corner detector to the case of color images. As we use for matching only first order derivatives of images, we need also here precise first order point of interest. We generalize here the precise Harris [7], [1] detector to color data.
let:
stands for a Gaussian smoothing with a standard deviation of 
.
The Harris color operator can then be define by positive local extrema of the following operator:
with: K=.04
The figure 7 represents two images differing from viewpoint and intensity. We have computed the Harris points on them. We can see that our detector is stable, according to the fact that most of the points obtained can be matched.
Figure 7: Harris points computed on two images differing from viewpoint and change of intensity.
These two images contain respectively 1924 (l: left image) and 3021
(r: right image) points.
These points have been obtained by computing derivatives at half pixel precision
and maximizing the resulting Harris Color Image
in a circular window of diameter of 11 pixels.