Object and Scene Recognition Based on Learning Probabilistic Latent Component Tree with Boosted Features
Masayasu Atsumi · International Journal of Machine Learning and Computing · 2012
This paper proposes an object and scene categorization method based on the probabilistic latent component tree with boosted features.In this method, object classes are firstly obtained by clustering a set of object segments extracted from scene images in each scene category through the probabilistic latent component analysis with the variable number of classes.Then the probabilistic latent component tree with boosted features at its branch nodes is generated as a classification tree of all the object classes of all the scene categories followed by labeling object classes.Lastly, each scene category is characterized according to the composition of its labeled object classes.Object and scene recognition is simultaneously performed based on the probabilistic latent component tree search by using composite boosted features for the tree traversal.Through experiments by using images of plural categories in an image database, it is shown that performance of object and scene recognition is high and improved by using composite boosted features in the probabilistic latent component tree search.