Mean representation based classifier with its applications

Based on a fundamental concept that most similar properties of samples from a single-object class should be congregated on their class mean, an efficient and simple approach for pattern identification, called the mean representation based classifier (MRC), is presented. MRC is a linear model representing a testing sample as a linear combination of all class means and the class associating the biggest item of the linear combination coefficient is favoured. MRC is easy to employ with a least squares estimator. In addition, MRC need not tune any parameter and avoids mistaking the local optimum value as the global optimal one. MRC is evaluated on three standard databases. The experimental results show MRC is superior to other state-of-the-art nonparametric classifiers.