An Effective Machine Learning Approach for Refining the Labels of Web Facial Images
Jieh-Ren Chang, Hung-Chi Juang · 2015
he technique of search-based face annotation is implemented by mining weakly labeled facial images that are freely collected from the internet web sites but is incompletely correct label data. In this study, the particle swarm algorithm and binary particle swarm algorithm are used to achieve the technique of Unsupervised Label Refinement (ULR) for refining the labels of web facial images. The experimental data is provided from IMDb website and with 45% initial incorrect label mark rate. The results show that the particle swarm algorithm and binary particle swarm algorithm have the better correction rate and convergence performance than other approaches.