A PARALLEL, HIERARCHICAL SCENE CLASSIFICATION FRAMEWORK

摘要

A novel parallel, hierarchical scene classificationrnframework is presented in this paper. At first, we use anrnimage pyramid to present both the global scene and thernlocal patches containing specific objects. Secondly, wernbuild our own codebooks, which extract both global andrnlocal features. Next, we train the visual words byrngenerative and discriminative methods respectively,rnwhich could obtain the initial scene categories based onrnthe potential semantic that is reasoned by the bag-ofwordsrnmodel. Then, we use a neural network to obtain thernfinal scene categories. Experiments show that the parallel,rnhierarchical image representation and classification modelrncould obtain better results.