Two Statistical Filters for Detecting Changes in Color Images

摘要

Two spatial statistical filters for the detection of changes in two color images are presented. Both are implemented efficiently using sliding windows and result in F-distributed test statistics, which are used to test the null (no-change) hypothesis for each pixel in the image. The first method tests for no changes in the model parameters used to describe the statistics in the region of the image surrounding a given test pixel. The second method uses two concentric windows centered on the test pixel and tests for equal squared differences summed over the two (foreground and background)windows. The second method was found to be superior to the first. A single illustrative example is presented.