Research Of Fast Face Retrieval Algorithms In Large Scale Database

Face recognition as a biometric identification technology, has been widespread concern in recent years and also has become a hot issues with the closely combination of applied mathematics and information technology.Although face recognition has been used in many ways, but under the large scale database, traditional search method searching face one by one could not satisfy the real time requirement. Aiming at solving the face retrieval problem in large scale database, we proposed a new fast face retrieval algorithm based on k-means clustering and doing recognition by calculating partial feature at different levels. In our algorithm:We first applied our modified LBP coding method to extract features of face pictures from the database and generate the database of features, and then we assigned each part with different weight according to their recognition abilities.Next we did the clustering operation utilizing the improved K-means algorithm on the former feature database and we reserved the indexes of different clusters.Then we do face retrieval by our fast search algorithm using re-rank method to improve the result.Finally we experiment on the database collected by ourselves, experimental results showed that our received a better face retrieval system by obviously increasing the retrieval efficiency with the precision descending less than two percents.