Image context classification based on visual codebook feature boosting

This paper presents a method for classifying the context of images. The context of an image can be classified as indoor, outdoor or a more specific scene category. Several state of the art methods use visual codebooks in order to construct global image descriptors and classify the latter using a Support Vector Machine (SVM) classifier. This paper proposes boosting over visual codebook features as an alternative to SVM classification. The boosting based approach has several advantages: fast training and classification time, no need for classifier parameter tuning, efficient combination of different descriptor types, small classifier models. The proposed method performs well on large datasets with many classes and provides state of the art results.