Correspondence analysis and hierarchical indexing for content-based image retrieval

Abstract

This paper describes a two-stage statistical approach supporting content-based search in image databases. The first stage performs correspondence analysis, a factor analysis method transforming image attributes into a reduced-size, uncorrelated factor space. The second stage performs ascendant hierarchical classification, an iterative clustering method which constructs a hierarchical index structure for the images of the database. Experimental results supporting the applicability of both techniques to data sets of heterogeneous images are reported.

Nonparametric motion characterization using causal probabilistic models for video indexing and retrieval

“…As a consequence, to cope with video databases involving various dynamic contents, it is necessary to determine an optimal set of features and the associated similarity measure. These issues can be tackled using principal component analysis [31] or some other feature selection techniques [29]. Unfortunately, the feature space is usually of high dimension and the distance metric used is likely not to properly capture the uncertainty attached to feature measurements.…”

“…Considering an approximation of the KL divergence detailed in the Appendix and using the exponential form (12) of the likelihood function , the KL divergence of law w.r.t. law can be approximated by (31) Expression (31) quantifies the loss of information occurring when considering instead of to model the motion distribution attached to .…”

“…The general idea is to define an appropriate similarity measure between image sequences and to determine the closest matches according to this similarity measure. As far as feature-based techniques are concerned, the retrieval process generally makes use of classical distances in the feature space such as the Euclidean or Mahanalobis distances, [30], [31]. In our case, we first benefit from our statistical modeling of motion activity to define an appropriate similarity measure w.r.t.…”

“…We focus here on hierarchical representations that have been successfully exploited for browsing or retrieval in still image database [12], [31], [40], [45]. Such indexing structures rely on binary trees.…”

“…The tree nodes will correspond to subsets of shots of the processed video database. To achieve this hierarchical structuring, either top-down [40] or bottom-up [31] strategies can be adopted. As pointed out in [11], bottom-up techniques seem to offer better performance in terms of classification accuracy.…”