IDENTIFYING CRITICAL FEATURES IN ORDERED SCALE SPACE

摘要

A system and method for identifying critical features in an ordered scalespace within amulti-dimensional feature space is described. Features are extracted from aplurality of datacollections. Each data collection is characterized by a collection of featuressemantically-relatedby a grammar. Each feature is normalized and frequencies of occurrence and co-occurrences forthe feature for each of the data collections is determined. The occurrencefrequencies and the co-occurrencefrequencies for each of the features are mapped into a set of patterns ofoccurrencefrequencies and a set of patterns of co-occurrence frequencies. The patternfor each datacollection is selected and distance (similarity) measures between eachoccurrence frequency inthe selected pattern is calculated. The occurrence frequencies are projectedonto a one-dimensionaldocument signal in order of relative decreasing similarity using thesimilaritymeasures. Wavelet and scaling coefficients are derived from the one-dimensional documentsignal using multiresolution analysis.