Unsupervised face recognition by associative chaining

摘要: Abstract We propose a novel method for unsupervised face recognition from time-varying sequences of images obtained in real-world environments. The utilizes the higher level sensory variation contained input image to autonomously organize data an incrementally built graph structure, without relying on category-specific information provided advance. This is achieved by “chaining” together similar views across spatio-temporal representations space two types connecting edges depending local measures similarity. Experiments with gathered over period several months and including both frontal side-view faces 17 different subjects were used test method, achieving correct self-organization rate 88.6%. proposed can be video surveillance systems or content-based retrieval.