Unified image retrieval and keypoint matching by local geometric consistency and non-linear diffusion

Feature-based image retrieval and feature matching have been used together in many applications, but they have been treated as two separate problems. We propose an unified approach which, for a query image, finds a set of candidate images together with feature matching results. By considering the local geometric consistency of neighboring features, we can find more and better feature matches even in challenging situations. Since the proposed forward/backward matching and non-linear diffusion run very efficiently, they can be used in the candidate image selection and improve the image retrieval performance significantly. Through quantitative comparisons we show that the proposed approach performs better than the recent state-of-the-art feature matching algorithms and image retrieval algorithms.

I. Introduction

Local features have been used in many application areas, including robotics, augmented reality, and 3D modeling of objects or scenes. Appearance and geometric properties of image contents are captured by feature locations and descriptors, and they are used in finding correspondences in images and reconstructing geometry of the scene. Many geometric modeling tasks, specifically SLAM (simultaneous localization and mapping) or structure from motion, require fast and robust matching of features between multiple images. In addition, feature-based location recognition (loop closure detection) or object recognition need finding the images similar to the query image from many database images. For successful applications it is crucial to be able to retrieve the best matching image in a large collection of images and find good feature matches between the matched images.