一般化学習ベクトル量子化による画像認識

抄録

This paper describes image recognition by using Generalized Learning Vector Quantization (GLVQ). GLVQ has been proposed as a learning method of reference vectors that ensures convergence of them during learning. By formulating a novel learning scheme called General Loss Minimization (GLM) based on Bayes decision theory, GLVQ can be extended to various discrimination methods including Inverse of Lorentzian Mixture (ILM) and Discriminative Dimensionality Reduction (DDR). Application to motorcycle recognition reveals that the proposed method work well in the real world, and it contributes to the achievement of safe and secure societies.