分块PCA(Modular Principal Component Analysis)人脸识别算法在模式识别领域得到了广泛应用,但是该方法在识别表情和光照变化强烈的人脸方面存在不足。针对上述缺陷,本文提出了一种改进的基于分块加权的识别方法。该方法首先对所有分块分别进行匹配,然后通过归一化调低变化较大的图像块的权值并对匹配结果进行加权融合,从而削弱人脸局部变化过大产生的干扰。同时,引入了GPU加速技术,对该方法进行了加速。实验结果表明,该方法的识别精度明显优于传统的PCA方法及分块PCA方法,速度也可得到极大提升。
【英文论文摘要】
Modular Principal Component Analysis(Modular PCA) technique is widely used in human face recognition,but it fails when human face expression or illumination changes intensively.In this paper,an improved Modular PCA technique is presented.Firstly,image-modules are matched separately.Then the weights of the modules are set through distance normalization to weaken the scope noise.Finally,a match result is got after the fusion of the weighted modules.Moreover, the algorithm is speeded up on CUDA platform based ...