Entropy change detection
GEOMATRIX UAB has automated a highly sensitive and efficient change detection method presented in on-line publication by a group of researchers from Federal Institute for Geosciences and Natural Resources (BGR). The method is based on assumption that sudden increases of entropy on a panchromatic image reflect appearance of certain objects in plain (i.e. "uniform") surfaces. While comparing (by subtraction) entropies of two images taken with sufficient period of time, changes in entropy clearly represent the areas where changes have occurred. As a mathematical function, entropy is used to measure the level of "disorder" within a certain sample of values - practically speaking, a certain entropy value is assigned to a pixel by measuring entropy of a sample of pixel values present in a window around that pixel. Entropy difference masks can be created in two different ways in order to extract two different categories of changes: (1) subtraction of earlier image entropy from the later image entropy indicates changes related to features that occurred and (2) subtraction of later image entropy from the earlier image entropy indicates changes related to features that disappeared during the time span between two dates. There are several advantages of this change detection method: (1) it only demands panchromatic images with the same pixel size and data depth, which means that any VHR satellite, aerial photo or even UAV imagery can be used after a few automated pre-processing steps; (2) change detection method has high sensitivity and allows automated classification of change types. The method is highly demanding for processing power and RAM but overall processing efficiency can be significantly improved by automated parallel processing. There are many obvious areas of use for entropy change detection in urban and agricultural applications, as well as dual-use intelligence gathering.
Examples on the left show a real case of entropy change detection if the suburban area of Warsaw: top left - fragment of panchromatic Ikonos (1m) image taken in 2002; top-right - it's entropy calculated using 9 m moving window; mid left - panchromatic Ikonos (1m) image taken in 2008; mid right - it's entropy; bottom - entropy change mask in colors intensity representing change numeric values overlaid with the second (2008) Ikonos image. Green-yellow tones indicate natural succession process in abandoned fields, while red tones clearly show new buildings (including even those already under construction in 2002).