Probabilistic Visual Learning for Object Representation-英文文献.pdfVIP

1 Probabilistic Visual Learning for Object Representation Baback Moghaddam and Alex Pentland Massachusettes Institute of Technology Abstract We present an unsupervised technique for visual learning which is based on density estimation in highdimensional spaces using an eigenspace de composition Two types of density estimates are derived for modeling the training data a multivariate Gaussian for unimodal distributions and a MixtureofGaussians model for multimodal distributions These pr oba bility densities are then used to formulate a maximumlikeliho o d estima tion framework for visual search and target detection for automatic object recognition and coding Our learning technique is applied to the probabilis tic visual modeling detection recognition and coding of human fac es and nonrigid objects such as hands Introduction Visual attention is the pro cess of restricting higherlevel pro cessing to a subset of the visual eld referred to as the fo cusofattention FOA The critical comp o nent of visual attention is the selection of the F OA In humans this pro cess is not based purely on b ottomup pro cessing and is in fact goaldriven The measure of interest or saliency is mo dulated by the b ehavioral state and the demands of the particular visual task that is currently active Palmer has suggested that visual attention is the pro cess of lo cating the ob ject of interest and placing it in a canonical or ob jectcentered reference frame suitable for recognition or template matching We have develop ed a computa tional technique for automatic ob ject recognition which is in accordance with Palmers mo del of visual attention see section The syste