Image difference detection under varying illumination based on vector space and correlations

Abstract We propose two methods to detect differences in images independently of local changes of intensity. The methods are based on calculating geometrical operators when images are considered as vectors. Operators can be expressed in terms of correlations for the possibility of optical implementations. The methods are invariant to changes of the form af ( x , y ) +  b , where a and b are arbitrary unknown parameters that may vary over the image f ( x , y ). Computer simulations show that the method works well when the illumination model is satisfied. Results from real images taken with a web camera show the robustness of the method.