This paper investigates a framework of search-based face annotation by mining weakly labeled facial pictures that are freely offered on world Wide internet (WWW). One difficult drawback for search-based face annotation theme is a way to effectively perform annotation by exploiting the list of most similar facial pictures and their weak labels that are usually clamorous and incomplete. To tackle this drawback, we have a tendency to propose an efficient unsupervised label refinement (ULR) approach for processing the labels of internet facial pictures exploitation machine learning techniques. We have a tendency to formulate the training drawback as a plan convex improvement and develop effective improvement algorithms to resolve the large-scale learning task expeditiously. To more speed up the planned theme, we have a tendency to additionally propose a clustering-based approximation algorithmic program which might improve the scalability significantly.
