Correlation-based registration of bland seafloors for coherent change detection

To perform coherent change detection between a pair of synthetic aperture sonar images, they must be registered to within a fraction of a pixel. The act of registering two bland seafloor images has been described as the `worst-case scenario' due to the lack of features in the scene to act as markers. However, the speckle noise that is common to all coherent imaging is a deterministic process caused by the sub-resolution scatterers in the scene. This implies that the speckle can be used to register the images, provided that sources of decorrelation (e.g., baseline and temporal decorrelation) are sufficiently low. This paper details a study into registering bland imagery using correlation to measure the warping between two images and least squares methods to estimate the parameters of a simple track error model. Rotations between the images are shown to cause a significant decrease in correlation and must be corrected using a preliminary optimisation technique. Using simulated data, this approach is demonstrated to be capable of registering bland seafloor images to within the required one-tenth of a resolution cell, including in the presence of temporal decorrelation.