Statistical part-based models for object category recognition

In this paper, we present a new method to learn statistical part-based structure models for object category recognition in a supervised manner. The method learns both a model of local part appearance and a model of the spatial relations between those parts. By using histograms of oriented gradient (HOG) features to describe local part appearance within an image, we investigate whether richer appearance model is helpful in improving recognition performance. We learn the model parameters from training examples using maximum likelihood estimation. In detection, these models are used in a probabilistic way to classify and localize the objects in the images. The experimental results on a variety of categories demonstrate that our method provides both successful classification and localization of the object within the image.