Porcelain image classification based on semi-supervised mean shift clustering

Abstract

This paper puts forward a method of constructing efficient clustering model using sample constraints at the problem of porcelain image classification. The method is based on the mean-shift clustering algorithm, the mapping kernel transformation matrix is computed by using the sample constraint, and defined using the Intra-class consistency measure and Inter-class sample difference in distribution measurement, the method of determining the best clustering bandwidth parameters is achieved, and an effective semi-supervised clustering model is obtained. The experimental results show that compared to the common fuzzy mean clustering, kernel fuzzy mean clustering and some improved algorithms based on it, This method has been greatly improved in the clustering effect, the classification method is applied to the porcelain image, to achieve a more accurate classification with individual image tiles effectively.