Local vector space operators for detection of differences in images under varying illumination

In this paper we define operators based on local vector space representation for detecting changes in images when the illumination of the scenes varies. The difference image contains information from objects not present in original images. Most change image detection algorithms assume that the illumination of a scene will remain invariable. But when the illumination conditions of a scene cannot be controlled, image processing algorithms must be adapted. Here we propose methods based on a local vector model when the illumination of a scene may vary locally and when no knowledge about the illumination parameters is available. We evaluate three operators for image difference independently of local changes of intensity. In addition, all difference local operators are defined in terms of correlations which can be useful for optical implementations using conventional Vander Lugt or joint transform correlators.