Mobile landmark recognition

Abstract

Significant advances have been made in the field of computer vision, in particular Mobile Visual Search and the recognition of objects and places virtually through the development of feature detectors and descriptors. This project compares the performance of three key point detectors, Difference-of-Gaussians (DoG), Harris-Affine and Hessian-Affine detectors, based on repeatability, in the presence of image transformations, such as changes in viewpoint angle, scale, image blur, JPEG compression and illumination. The two more common matching algorithms, Scale-Invariant Feature Transform (SIFT) and Speeded-Up Robust Features (SURF), are also compared in terms of matching and recognising the same scene using the same image data set. The experiment has found the Hessian-Affine detector and SURF to have the best performance.