Abstract :
In this paper, Spatial Relationship model is presented as a novel technique of learning spatial models for visual object recognition. In contrast to other methods which explicitly give some parameterized spatial models, the proposed algorithm uses a latent class model to reveal some certain latent spatial relations. The advantages of the proposed model include: (1) it uses an unsupervised learning paradigm which can avoid some manual controls; (2) it can obtain some translation, rotation, scale and affine invariant properties; (3) The spatial relationship is latent which perhaps has more insight into describing the object structure. Combined SR with statistical visual word, SR-S is developed as an implementation of object recognition algorithm. SR-S uses an unsupervised process that can capture both spatial relations and visual word appearances simultaneously. The experiments are demonstrated on some standard databases and show that SR is a promising model for analysing object spatial relationship.
