From variant index images (like NDVI) derived from time-serial remote sensing images, both the ‘spatial variation’ and ‘temporal variation’ for multiple disciplines could be defined, which are widely available, in particular, in the literatures. However, most of those research results show high uncertainty and low reliability.
In practice, in order to *accurately* define spatial variations or detect spatial changes directly from time-serial images or index images, many technical challenges are required to be solved.
For example, in GIS and land management, how *effectively* to detect spatial changes of landcover over time (i.e., landuses, building lots, fences, tree crowns, etc.)?
Obviously, algorithms for change detection /defining spatial variation are mostly different from feature extraction, including traditional change detection algorithms and object-oriented algorithms.
For last few years, many researches and practitioners have been discussing object-oriented Change Detection with eCognition and ERDAS Objective.
With eCognition, it uses the multivariate alteration detection (MAD) transformation (by Allan et al, 1998; Nielsen & Conradsen, 1997), which is based on the established canonical correlations analysis.
However, it looks that MAD might be challenging, as a ‘real' object-oriented solution for Change Detection in accurate way.
Inversely, ERDAS Objective uses Discriminant Function Change algorithm to help extract change features, which demonstrates direct and efficient way to map ‘spatial variation’ and perform change detection...
