Automatic Geometric and Radiometric Registration of Landsat-TM Images Using Mutual Information

This work is on development of a method for automatic registration of satellite images acquired on different dates, for both geometric and radiometric correction with respect to a reference image. Mutual information statistics is used as the similarity metric of geometric and radiometric registration. Affine and linear transformations are used in geometric and radiometric correction respectively. Powell's method is applied in iterative optimization to find the best transformation parameters for both types of registration, based on the maximum mutual information between images. The method is validated using Landsat's Thematic Mapper (TM) sensor, bands 3, 4 and 5 images, obtained on five separate dates for scene 231-062 in the Central Amazon.