Rotation and Size Independent Face Recognition by the Spreading Associative Neural Network
Keisuke Nakamura, Hironobu Takano · The 2006 IEEE International Joint Conference on Neural Network Proceedings · 2006
Emulating the parietal cortex, a "rotation and size spreading associative neural network" (RS-SAN net) was developed. Using the RS-SAN net, a new personal authentication method was proposed which was not influenced by the rotation (in plane) and size changes of the input faces. The recognition characteristics of the RS-SAN net for both learned (familiar) and un-learned (unfamiliar) face images were investigated in various plane rotations and sizes. The RS-SAN net had fairly good orientation and size recognition characteristics only for learned faces, but not for unlearned faces. Thus, the orientation and size of the input face image were rightly corrected only for the learned faces. By adding the inner product and minimum distance as new shape recognition criteria for the RS-SAN net, both the learned and unlearned face images were recognized correctly. After the RS-SAN net corrected both the orientation and size to the registered ones for the arbitrary rotation and size of the input faces, both the false acceptance and false rejection rates were 0% in appropriate threshold ranges; the equal error rates became 0%.