What Is a Good Nearest Neighbors Algorithm for Finding Similar Patches in Images?

Abstract

Abstract. Many computer vision algorithms require searching a set of images for similar patches, which is a very expensive operation. In this work, we compare and evaluate a number of nearest neighbors algorithms for speeding up this task. Since image patches follow very different distributions from the uniform and Gaussian distributions that are typically used to evaluate nearest neighbors methods, we determine the method with the best performance via extensive experimentation on real images. Furthermore, we take advantage of the inherent structure and properties of images to achieve highly efficient implementations of these algorithms. Our results indicate that vantage point trees, which are not well known in the vision community, generally offer the best performance.

Semi-supervised hashing for scalable image retrieval

“…Over the past decade, several approximate nearest neighbor (ANN) search techniques have been developed for large scale applications. Although there exist many tree-based methods e.g., [3,1,11,14,10], for applications with memory constraints, hashing based ANN techniques have attracted more attention. They have constant query time and also substantially reduced storage as they usually store only compact binary codes for each point in X .…”

Parallel k nearest neighbor graph construction using tree-based data structures

“…Moore et al uses the idea of anchors instead of balls and uses the triangle inequality to efficiently build a ball tree that prunes nodes which would not belong to the current child [11]. Kumar et al do a comprehensive survey of tree-based algorithms for nearest neighbor search [8]. Multiple, randomized k-d trees (a k-d forest) are proposed in [19] as a means to speed up the approximate nearest neighbor search; this is one of the most effective methods for matching high dimensional data [12].…”

Mixed-Resolution Patch-Matching

“…It is practical only for local matching and has an increased memory overhead. Many other methods for finding approximate or exact nearest patches have been suggested, a review of which can be found in [13].…”