Superpixel-Based Feature Tracking for Structure from Motion

Feature tracking in image collections significantly affects the efficiency and accuracy of Structure from Motion (SFM). Insufficient correspondences may result in disconnected structures and incomplete components, while the redundant correspondences containing incorrect ones may yield to folded and superimposed structures. In this paper, we present a Superpixel-based feature tracking method for structure from motion. In the proposed method, we first propose to use a joint approach to detect local keypoints and compute descriptors. Second, the superpixel-based approach is used to generate labels for the input image. Third, we combine the Speed Up Robust Feature and binary test in the generated label regions to produce a set of combined descriptors for the detected keypoints. Fourth, the locality-sensitive hash (LSH)-based k nearest neighboring matching (KNN) is utilized to produce feature correspondences, and then the ratio test approach is used to remove outliers from the previous matching collection. Finally, we conduct comprehensive experiments on several challenging benchmarking datasets including highly ambiguous and duplicated scenes. Experimental results show that the proposed method gets better performances with respect to the state of the art methods.

Bi-Level Locality Sensitive Hashing Index Based on Clustering

Locality sensitive hashing is the most popular algorithm for approximate nearest neighbor search. As LSH partitions vector space uniformly and the distribution of vectors is usually non-uniform, it poorly fits real dataset and has limited search performance. In this paper, we propose a new Bi-level locality sensitive hashing algorithm, which has two-level structures to perform approximate nearest neighbor search in high dimensional spaces. In the first level, we train a number of cluster centers, then use the cluster centers to divide the dataset into many clusters and the vectors in each cluster has near uniform distribution. In the second level, we construct locality sensitive hashing tables for each cluster. Given a query, we determine a few clusters that it belongs to with high probability, and then perform approximate nearest neighbor search in the corresponding locality sensitive hash tables. Experimental results on the dataset of 1,000,000 vectors show that the search speed can be increased by 48 times compared to Euclidean locality sensitive hashing, while keeping high search precision.