Automated registration for remote sensing images using PCNN and invariant moments approaches

Abstract EN

Digital image registration is very important in many applications such as medical image analysis, robotics, and remote sensing.

Image registration determines the relative orientation between two images.

The present work represents two feature-based approaches to automated image-to-image registration.

The characteristic of both approaches is that it uses the Harris detector for identification of interest points in both images; the first approach uses the pulse coupled neural network (PCNN) signature to establish correspondences between the potentially matched regions defined around detected points from the two images, while the second approach uses the invariant moments instead of using PCNN signature for matching.

The two algorithms were tested on Landsat Thematic Mapper (TM) and Spot remote sensing images showing that the Root Mean Square Error (RMSE) for PCNN approach is less than the moment approach while the moment technique has the lowest running time.