Study of Change Detection based on Edge Detection of Satellite Images

This paper focuses on a robust and flexible edge detection technique based on Independent Component Analysis (ICA) on satellite images for recognition and subjective analysis of changes caused due to Japan earthquake 2011 followed by a devastating tsunami. ICA has been applied to satellite image patches to learn the basis functions using the fixed-point FastICA algorithm. As most of the basis functions are sparse, they are used as pattern template for feature extraction. These basis functions are usually localized, band-limited and oriented like Human Visual System (HVS) and resemble as Gabor wavelet basis function. In the proposed edge detection technique GeoEye’s IKONOS satellite images corresponding to pre and post events have been first transformed to pattern maps (feature map) in which edges and background pixels have been classified into different classes. The edges have been extracted by using sparse components only, whereas non sparse components have been suppressed and treated as background.