Defence Electronics Applications Laboratoty, Dehradun

Abstract

A system has been built and tested for automated change detection between multi-temporal panchromatic images. This paper discusses the implementation issues, associated tools, and finally summarises initial tests on IRS IC/ID and other high-resolution images. Key characteristics of this system are integration of technologies having high degree of registration, normalisation of the effects of radiometry; selectivity to specific type of changes, refinement of changes by thresholding, and assignment of presence and absence of object and tools for updation/deletion of change mask. A semi-automatic technique for selection of control points in an image having affine distortion has been implemented. Linear regression is used for normalisation of the images. Two change detection techniques, namely image subtraction and image ratioing have been used to find the global change mask. Selective threshold is used to generate target mask. Target mask is shown in two colours to depict presence and absence of the object. Method based on ratioing has been found to be more sensitive to spectral variations and provides better detection of changes.