Scalable face image retrieval integrating multi-feature quantization and constrained reference re-ranking

Scalable face image retrieval is an important and challenging issue in many real world applications such as security and surveillance systems. However, the current available face recognition approaches are facing the challenge of handling large-scale datasets due to their high dimensionality of features. Meanwhile, most of the proposed retrieval methods fail to take full advantage of useful information such as age and gender. To address the above issues and build a practical face image retrieval system, this paper proposes a multi-feature quantization method to quantize both component-based and global geometric features, based on which the candidate images can be discriminated. Afterwards, a re-ranking algorithm constrained by the classification performance is proposed to select the top ranked images as the final retrieval results. Experimental results have been illustrated and analyzed to show that the proposed methods outperform the state-of-art methods and achieve good face image retrieval results.