Scale-rotation invariant features from Non-Subsampled Contourlets

This work aims to build scale and rotation invariant features from Non-Subsampled Contourlet Transform (NSCT). The features will have properties similar to the popular Scale Invariant Feature Transform (SIFT). The features will be theoretically (and practically) invariant to scale, location and rotation. We also take care that practically they are invariant to changes in illumination as well. Our scale invariant features can be applied virtually anywhere SIFT features had been employed previously - object recognition, object detection, panorama etc. In this paper, we will show two examples how the features may be used for object recognition and for image stitching.