Incorporating visual adjectives for image classification

Image classification is a fundamental problem in computer vision which implies a wide range of real-world applications. Conventional approaches for image classification often involve image description and training/testing phases. The Bag-of-Features (BoF) model is one of the most popular algorithms for image description, in which local descriptors are extracted, quantized, and summarized into global image representation.In the BoF model, all the visual descriptors are naturally treated as nouns, and plenty of useful contents are ignored. In this paper, we suggest to extract descriptive information, known as adjectives, to help visual recognition. We propose a simple framework to integrate various types of adjectives, i.e., color (or brightness), shape and location, for more powerful image representation. Experimental results on both scene recognition and fine-grained object recognition reveal that our approach achieves superior classification accuracy with reasonable computational overheads. It is also possible to generalize our model to many other multimedia applications such as large-scale image search.