Dual threshold based unsupervised face image clustering

Face image clustering is a very useful technique that can be used in photo classification, face image retrieval, etc. Under many circumstances, the number of clusters is unknown and the estimation of the number is a complicated problem. To cluster face images without cluster number as a prior, a novel unsupervised face image clustering algorithm is proposed based on LGBPHS model and threshold. We cluster the face images which are not linearly discriminated by employing dual threshold strategy. By merging data iteratively with proper thresholds, the algorithm automatically produces cluster number and clusters. Experimental results demonstrate our algorithm performs well in Yale data, and the dual threshold obtained from Yale data generalizes well to ORL data.