Biology vision inspired singularity model in invariant recognition

Invariant recognition is a traditional challenge in computer vision. A biology vision inspired model is proposed to realize rotation invariant recognition. Neurobiological plausibility of the model is expressed in three aspects: Gabor filters pair like complex cell, singularities and memory trace. Recurrent connections decrease distinction of complex cells leading to emergence of singularities. Memory trace extracts correlations of different views of the same objects from continual sequences, and therefore is fit for performing recognition tasks. We testify efficacy of the model by benchmark recognition problem.