Abstract:

We propose an improved local appearance and color modeling method, as an extension of Moghaddam & Zhou cite{BabackZhou}, for object detection and recognition in cluttered scenes. The approach is based on the joint distribution of local feature vectors at multiple salient points and factorization with Independent Component Analysis (ICA). We we are able to obtain a tractable set of joint probability densities which can model high-order dependencies in local image features. In this work we replace multi-dimensional histograms with Gaussian mixture models with model-order selection based on the Minimum Description Length (MDL) criterion. Furthermore, a hybrid color/appearance modeling scheme is introduced which significantly increases performance.