基于曲波变换和Retinex人脸光照处理算法

针对光照对人脸识别影响问题,提出了一种基于曲波变换和Retinex人脸光照处理的算法。该算法对光照变化人脸对数变化后的图片进行曲波变换(Curvelet);利用Kimmel变分模型作为平滑滤波算子对低频图像进行平滑滤波,对高频系数进行阈值去噪。通过曲波逆变换得到光照亮度成分图像;利用Retinex模型提取光照不变成分。通过Yale B与CMU PIE人脸库的实验结果表明:该算法能有效地消除光照变化对人脸识别的影响并提高人脸识别率。

Illumination processing method based on Curvelet transform and Retinex

An illumination processing method based on Curvelet transform and Retinex model is proposed to the problem of the illumination impact on the face recognition. The method performs Curvelet transform in the logarithm domain of the illumination variation face images. The Kimmel variational method is used as smoothing filter operator to smooth the low-frequency image. The threshold denoising is used in the high-frequency coefficients. The illumination brightness composition can be obtained by inverse Curvelet transform. The Retinex model is used for illumination invariant extraction. Experimental results from Yale B and CMU PIE databases show that proposed method can effectively eliminate the effect of varying illumination on face recognition and improve the rate of face recognition.