无训练样本的伪画像生成及识别

Face pseudo-sketch synthesis and recognition without training samples

基于照片的伪画像生成及识别是人脸识别领域的一个新研究方向,在刑侦方面有着重要用途。但目前的方法都需要大量的训练样本以生成伪画像,对此提出将双边滤波器用于提取人脸的纹理及形状特征,并利用颜色恒常性对重要特征区域进行增强,基于香港大中文学人脸画像(Chinese University of Hongkong face sketch,CUFS)数据库的实验结果表明,利用该方法生成的伪画像能有效提高画像的识别性能,是一种可行的伪画像生成方法。

Pseudo-sketch synthesis and recognition is a new direction of face recognition, which is important in forensics. However, most current methods need lots of training samples, so the bilateral filter is used to ex tract shape and texture feature of face, and the important feature is improvd with color constancy. Experiment based on Chinese University of Hongkong face sketch （CUFS） database illustrates that the proposed algorithm can efficiently improve the rate of recognition.