Issue No. 04 - April (2008 vol. 30)
We investigate whether dimensionality reduction using a latent generative model is beneficial forthe task of weakly supervised scene classification. In detail we are given a set of labelled images ofscenes (e.g. coast, forest, city, river, etc) and our objective is to classify a new image into one ofthese categories. Our approach consists of first discovering latent "topics" using probabilistic LatentSemantic Analysis (pLSA), a generative model from the statistical text literature here applied to a bagof visual words representation for each image, and subsequently training a multi-way classifier on thetopic distribution vector for each image. We compare this approach to that of representing each imageby a bag of visual words vector directly, and training a multi-way classifier on these vectors.To this end we introduce a novel vocabulary using dense colour SIFT descriptors, and then investigatethe classification performance under changes in the size of the visual vocabulary, the number oflatent topics learnt, and the type of discriminative classifier used (k-nearest neighbour or SVM). Weachieve superior classification performance to recent publications that have used a bag of visual wordrepresentation, in all cases using the authors' own datasets and testing protocols. We also investigatethe gain in adding spatial information. We show applications to image retrieval with relevance feedbackand to scene classification in videos.
Scene Classification, pLSA, Spatial Information
A. Bosch, A. Zisserman and X. Muñoz, "Scene Classification Using a Hybrid Generative/Discriminative Approach," in IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 30, no. , pp. 712-727, 2007.