Modeling multimedia contents through probabilistic feature signatures

Abstract

We introduce a new family of flexible feature representations for content-based multimedia retrieval: probabilistic feature signatures. While conventional feature histograms and feature signatures aggregate the multimedia objects' feature distributions exhibited in some feature space according to a partitioning, probabilistic feature signatures model these feature distributions by means of discrete or continuous probability distributions. In this way, they combine the advantages of high expressiveness and compactness, for instance through Gaussian mixture models. In this paper, we introduce the concept of probabilistic feature signatures and provide the empirical evidence of high retrieval performance when using this feature representation type. We show that probabilistic feature signatures are able to outperform conventional feature signatures.

Similarity query processing for probabilistic sets

“…6 We obtain more than 23 million annotations from the original dataset by removing users and urls that occurred less than ten times. We then map each tag to a set of urls to which the tag was used as part of a user's annotation.…”

“…A related work in multimedia retrieval is the recent Probabilistic signature method [6], where the set of feature points of an object are modeled as a distribution in the entire feature space. The distribution can be concisely described as a set of overlapping Gaussian distributions and efficiently learned via the EM algorithm.…”

Multimedia Data Mining and Processing for News Source Attribution

“…Probabilistic modelling is a method that statistically uses the effect of random incidences or activities in predicting the chance of future results [54]. These models are machine learning techniques purposely for predictions based on the important principles of statistics and probability.…”