Object Categorization based on Probabilistic Integration of Local and Global Features

T. Tanaka, K. Hotta, and H. Takahashi (Japan)


local features, global features, object categorize, probabilistic integration.


In conventional methods for object categorization, the background information produces unfavorable results. In general, there is a relationship between objects and back ground. If we can use the background information in object categorization effectively, the accuracy is improved more. In this paper, we use global features obtained from the entire image which contains the background information as well as local features. By using both types of features in categorization, it is possible to use background information effectively, as opposed to conventional methods, where the use of background information produces unfavorable results. We train the classifiers independently on the basis of global and local features, and the resulting classifiers are integrated by applying the Bayes theorem. The effectiveness of the proposed method is demonstrated in experiments using the Caltech 101 database.

Important Links:

Go Back