MINING WEAKLY LABELED WEB FACIAL IMAGES FOR SEARCH-BASED FACE ANNOTATION

Abstract

This paper investigates a framework of the search-based totally face annotation by mining weakly labeled facial images that are freely available on the world wide web. the one difficult trouble for the search-based face annotation scheme is how to efficiently perform annotation by means of exploiting the list of most comparable facial images and their weak labels which might be regularly noisy and incomplete. Tackleof this trouble, we recommend an powerful unsupervised label refinement method for refining the labels of web facial images the use of machine learning strategies. We formulate the learning trouble as convex optimization and develop powerful optimization algorithms to remedy the huge-scale learning task efficiently. further speed up the proposed scheme, we additionally recommend a clustering-primarily based approximation algorithm which can enhance the scalability extensively. We have got carried out an intensive set of empirical researches on a huge-scale web facial image test bed, in which encouraging outcomes showed that the proposed ULRalgorithms can considerably enhance the overall performance of the promising SBFA scheme.