Bayesian Kernel Inference for 2D Objects Recognition Based on Normalized Curvature
Kah Bin Lim, Wei Miao Yu, Tiehua Du · 2006
In this paper, we introduce the Bayesian kernel inference to the classical problem of 2D objects recognition using the normalized curvature as the feature vector. The idea and formulations of curvature normalization are proposed. A Gaussian window is applied to the feature before the normalization. The proposed Bayesian classifier in the hyperspace is formed by dramatically few kernels, which is significant for the problem with big learning sample database. The experiments show that this method has an excellent accuracy and insensitive to the noise. This algorithm also could easily be applied to multi-class problems.