Automatic registration of images with simulated rotation and translation estimation using HAIRIS

Abstract

Image registration is the most important fundamental phenomenon in the image processing system. In which Automatic image registration is a challenging aspect. Although enormous methods for automatic image registration have been developed and implemented in ancient days, it is still a broad use in plenty applications, such as in remote sensing. In my work, I have proposed a method for automatic image registration through histogram-based image segmentation (HAIRIS). This new approach is designed by combining several segmentations of the pair of images to be registered, according to a relaxation parameter based on the delineating histogram modes, following the characterization of the objects extracted - via the objects area, axis ratio, perimeter and fractal dimension - and a statistical procedure for objects matching is applied to each object. Finally, the simulated rotation and translation are illustrated for this proposed methodology. The first and foremost dataset consists of a photograph and a rotated and shifted version of this photograph is developed, with different levels of added Gaussian white noise. This can also be applied to satellite images which are in pair, with different spectral content and simulated translation and rotation is estimated, and also for various remote sensing applications comprising of different viewing angles, different acquisition dates and different sensors. Histogram-based image segmentation allows the registration of pairs of multitemporal and multisensor images with their differences in rotation and translation parameters, with small spectral content differences whereas, leading to sub pixel accuracy.