Represent and Detect Geons by Joint Statistics of Steerable Pyramid Decomposition

We present a parametric method to represent and detect geons. The parameters are extracted from joint statistical constraints de ned on complex wavelet transform. We rst review how steerable pyramid may be used in multi-scale multiorientation image decomposition. Then, four stages of object recognition theory is adopted to support the choice of joint statistical constraints, which characterize geons in a high dimensional parameter space. This parametric representation is examined in detail under circumstances when geons change in orientation, location, and size. Constraint-wise similarity is introduced to describe the corresponding statistics variations. Finally, we present details of a geon detection system, and demonstrate successful experimental results as well as system limitations.