Kernel optimization strategy based on mean shift

A kernel optimization strategy based on mean shift was proposed to improve the classification performance of samples. After doing mean shift optimization, the samples move to the center of local high density and the distances of samples in the same class are closer than the distances of the samples in different classes. We design an iterated local search strategy for automatic bandwidth selection of mean shift optimization. In experiments, we demonstrated the performance on various data sets. According to our experiment study, our method can improve the discrimination of samples.