Change detection using joint intensity histogram

In the present paper, a method for detecting changes between two images of the same scene taken at different times using their joint intensity histogram is proposed. First, the joint histogram, which is a two-dimensional (2D) histogram of combinatorial intensity levels, (I-1 (x), I-2 (x)), is calculated. By checking the characteristics of the ridges of clusters on the joint histogram, clusters that are expected to correspond to background are selected. The combinations of (I-1, I-2) covered by the clusters are determined as insignificant changes. Pixels having a different combinatorial intensity (I-1 (x), I-2 (x)) from these combinations, are extracted as candidates for significant changes. Based on the gradient correlation between the images for each region consisting of these pixels, only regions with significant changes are distinguished. Experiments using real scenes show the practical usefulness of the method.