The task of identifying the semantic localization of a robot has commonly been treated as a classification problem, where images are taken as input and a set of predefined labels is the output. While traditional approaches have focused on the performance of the image features extracted from computer vision techniques, the contextual information that can come with the images has not been taken into account. In this work, we present an approach for integrating this information in a scene classification pipeline where we opt for Bayesian network classifiers in addition to standard support vector machine ones. The approach is evaluated in two scenarios, one in which the contextual information is directly provided with the images, and the other where it must be inferred in an additional stage. The evaluation was performed using two families of classifiers over two datasets, and the results obtained show how the scene classification problem can benefit from the integration of contextual information