Spatial Pyramid Matching

This chapter deals with the problem of whole-image categorization. We may want to classify a photograph based on a high-level semantic attribute (e.g., indoor or outdoor), scene type (forest, street, office, etc.), or object category (car, face, etc.). Our philosophy is that such global image tasks can be approached in a holistic fashion: It should be possible to develop image representations that use low-level features to directly infer high-level semantic information about the scene without going through the intermediate step of segmenting the image into more “basic” semantic entities. For example, we should be able to recognize that an image contains a beach scene without first segmenting and identifying its separate components, such as sand, water, sky, or bathers. This philosophy is inspired by psychophysical and psychological evidence that people can recognize scenes by considering them in a “holistic” manner, while overlooking most of the details of the constituent objects (Oliva and Torralba, 2001). It has been shown that human subjects can perform high-level categorization tasks extremely rapidly and in the near absence of attention (Thorpe et al., 1996; Fei-Fei et al., 2002), which would most likely preclude any feedback or detailed analysis of individual parts of the scene. Renninger and Malik (2004) have proposed an orderless texture histogram model to replicate human performance on “pre-attentive” classification tasks. In the computer vision literature, more advanced orderless methods based on bags of features (Csurka et al., 2004) have recently demonstrated impressive levels of performance for image classification. These methods are simple and efficient, and they can be made robust to clutter, occlusion, viewpoint change, and even non-rigid deformations. Unfortunately, they completely disregard the spatial layout of the features in the image, and