Multi-view invariant shape recognition based on neural networks

Several shape recognition systems based on pairwise shape matching technique have achieved high accuracy but they face a problem of time consumption when they are evaluated on a large database. So this drawback makes the system impractical for real-time applications. Motivated by this obstacle, we have investigated a novel and robust neural network solution to achieve high speed of shape recognition without sacrificing accuracy via the non-absolute 1-D triangle area representation (NATA). Our method has been evaluated over a number of affine distorted shapes. The experimental results demonstrate that a shape recognition system using the neural network can achieve high speed and accuracy comparable with the prior system.