Query by Example in Large Databases Using Key-Sample Distance Transformation and Clustering

Abstract

Calculating the similarity estimates between the query sample and the database samples becomes an exhaustive task with large, usually continuously updated multimedia databases. In this paper, a fast and low complexity transformation from the original feature space into k-dimensional vector space and clustering are proposed to alleviate the problem. First k keysamples are chosen randomly from the database. These samples and a distance function specify the transformation from the series of feature vectors into k-dimensional vector space where database (re)clustering can be done fast with plurality of traditional clustering technique whenever required. In the experiments, similarity between the samples was calculated by using the Euclidean distance between their associated feature vector probability density functions. The k-means algorithm was used to cluster the transformed samples in the vector space. The experiments show that considerable time and computational savings are achieved while there is only a marginal drop in performance.

Analysis of acoustic-semantic relationship for diversely annotated real-world audio data

“…We use a dimensionality reduction method, by randomly selecting a number of q sound events from the data set that are used as anchor points indexed by a. Then we construct q dimensional feature vectors f n for each sound event n by measuring the distance E(X n , X a ) from the sound event n to the anchor points [16]. This translates into using q random columns of the similarity matrix S, which in practice means avoiding the calculation of the full similarity matrix.…”

On enabling techniques for personal audio content management

“…Thus, we have applied key sample transformation and clustering algorithm [7]. The transformation from series of feature vectors to a k-dimensional feature space is required in order to effectively cluster the database but at the same time minimum amount of information should be lost.…”

“…Efficient similarity measures for personal audio content management purposes have been recently proposed in [7] and [8].…”

Multiobjective Time Series Matching for Audio Classification and Retrieval

“…However, the choice of labels is still subjective and can result in conflicts on a semantic level with acoustically different sounds mapped to the same onomatopoeia. Some QBE systems use clustering before retrieval given that search time could be reduced by comparing the query only to a relevant cluster [34], [83]. However, building hierarchical classes implies that the database is created according to a specific dataset.…”