Home People Research Publications Demos
News Jobs Prospective
About Internal

Holistic Context Models: Classification Results

On Natural Scene Dataset

Scene classification can be implemented by assigning the query image to the scene class of largest posterior probability. Table.1 compares the resulting classification accuracies for N15, N13, and N8, with those of many methods in the literature.

Table 1

On Corel Dataset

Finally, Table.2 presents classification results for the C50 and C43 datasets. Contextual modeling again improves on the classification accuracy achievable with appearance classifiers. For C50 the absolute gain is of 4.2%, for C43 of 3%. Overall, it can be concluded that the proposed contextual modeling consistently outperforms existing context-based scene classification methods in the literature.