Spatial Neighborhood Mutual Information based Satellite Image Change Detection

Change detection deals with the problem of detecting changes that have occurred between various multi-temporal satellite images. This issue can be accomplished by measuring the similarity among these images. Therefore, in this paper, a very simple and effective change detection technique based on mutual information (MI), which is used as a similarity measure tool between two variables in statistics, is proposed. Herein, spatial neighborhood information around each pixel is exploited to get the MI and corresponding features. Further, difference feature vectors are created in feature space that provides discriminant information for change detection task. These difference feature vectors from all bands of multispectral images are concatenated to get the final feature vectors. Finally, features are classified by applying hard and soft clustering techniques i.e. k-means and fuzzy c-means, respectively and the results by both the clustering algorithms are compared. Experiments are conducted on two bitemporal satellite images, which confirm the effectiveness of the proposed technique.