Two-dimensional cellular neural networks for pre-processing in face recognition and digital library search

The architecture and the components design of a compact VLSI model based on the extension of the cellular neural network paradigm is described. Face and image recognition is an important function in future smart, portable multimedia systems. However, it involves many complicated mathematical operations that requires extremely large computing power. Since most operations are in two-dimensional format, the cellular neural network paradigm can be extended for selective tasks in face recognition function. Compact neural network is a suitable computing architecture. It uses the state-constrained neuron model that prevents the state variables becoming unbounded. The clear advantage is that the network will converge to proper solutions quickly. In addition, the hardware annealing technique can be applied to the compact neural network architecture so that the optimized solution can be quickly obtained without the drawback of conventional simulated annealing searching complexity and time penalty.