Skip to Main Content
An approach based on Bayesian framework and relevance feedback is proposed to improve the accuracy of indoor-outdoor image classification. In the system, knowledge from low-level features and spatial properties are integrated in Bayesian framework, and a relevance feedback method is implemented to specify the optimal weights of sub-blocks of images. The system provides the ability to utilize the local and spatial properties to classify new images. Performance testing of the algorithm is conducted using a database of real consumer photos. Experimental results over more than 1500 images show that high accuracy could be obtained using the spatial properties.
Machine Learning and Cybernetics, 2003 International Conference on (Volume:5 )
Date of Conference: 2-5 Nov. 2003