Image Retrieval Using Eigen Queries

Abstract

Abstract. Category based image search, where the goal is to retrieve images of a specific category from a large database, is becoming increasingly popular. In such a setting, the query is often a classifier. However, the complexity of the classifiers (often SVMs) used for this purpose hinders the use of such a solution in practice. Problem becomes paramount when the database is huge and/or the dimensionality of the feature representation is also very large. In this paper, we address this issue by proposing a novel method which decomposes the query classifier into set of known eigen queries. We use their precomputed results (or scores) for computing the ranked list corresponding to novel queries. We also propose an approximate algorithm which accesses only a fraction of the data to perform fast retrieval. Experiments on various datasets show that our method reports high accuracy and efficiency. Apart from retrieval, the proposed method can also be used to discover interesting new concepts from the given dataset.

Efficient Evaluation of SVM Classifiers Using Error Space Encoding

“…For example, one might apply principal component analysis (PCA) on query logs and fix W p as top eigen vectors [11]. Instead of PCA, other methods like linear discriminant analysis (LDA) [5] can also be used to achieve the same goal.…”

“…We find that ESE is 34 times faster than exhaustive SVM evaluation with comparable accuracy. In addition to ESE various pruning techniques like Subset Search [11] and Top-K Ranking [10], can be used to achieve high speedup. However, in this paper we focus only on the encoding scheme itself.…”

“…Next, we compare ESE with recent technique of approximation through eigen queries [11]. The method approximate the query using LSE, which performs poorly on rare queries (as shown in Section II-B).…”

“…On the other hand our method approximates the repeated queries with the nearest query in query logs without extra computations. We compare our method with [11] on Var20 dataset by executing 100 frequent and 100 rare queries. We vary the number of basis classifiers/eigen queries and compute average Recall@10 over all queries.…”

“…The number of classes to be considered in such a setting is still quite limited, and often known apriori. Raval et al [11] proposed a method to learn eigen queries using principal component analysis and used them for efficient image retrieval. However, this method fails to handle queries which are not in the span of the limited number of eigen vectors considered.…”